{
  "markdown": "# Coinrule MCP: Build AI Trading Agents Through Natural Language\n\n[![License: MIT](https://img.shields.io/badge/License-MIT-1F6FEB?style=flat-square)](LICENSE)\n[![MCP: Streamable HTTP](https://img.shields.io/badge/MCP-Streamable_HTTP-5A45FF?style=flat-square)](https://modelcontextprotocol.io/)\n[![OAuth 2.1](https://img.shields.io/badge/Auth-OAuth_2.1-0A7B83?style=flat-square)](https://cloud.coinrule.com/docs/mcp/connecting-your-ai-assistant)\n[![Works with Claude, ChatGPT, and Grok](https://img.shields.io/badge/Works_with-Claude_%7C_ChatGPT_%7C_Grok-111827?style=flat-square)](https://cloud.coinrule.com/docs/mcp)\n\n*Last updated July 22, 2026*\n\nAI is changing the way traders interact with financial markets. Instead of relying exclusively on dashboards, configuration panels, charts, and manually constructed automation, traders can increasingly use AI assistants as an intelligent interface between a trading idea and the infrastructure required to analyse, test, automate, and monitor it.\n\n**Coinrule MCP https://coinrule.com/mcp/** brings this conversational approach to automated trading.\n\nBy connecting a compatible AI assistant such as ChatGPT, Claude, Gemini, Grok, or another MCP-compatible client to a Coinrule account, traders can use natural-language instructions to inspect portfolios, analyse running strategies, explore trading ideas, perform historical backtests, create automated strategies, monitor signals, and manage trading agents.\n\nThe objective is not simply to add a chatbot to a trading platform. It is to create a more intelligent layer for interacting with algorithmic trading infrastructure.\n\nInstead of navigating through multiple interfaces to answer questions such as:\n\n> Which of my strategies generated the latest signal?\n\nor:\n\n> How is my portfolio currently allocated across connected venues?\n\na trader can ask an AI assistant directly.\n\nMore advanced workflows can go further:\n\n> Design a momentum strategy for BTC that only enters when the broader trend is positive, include defined risk controls, explain the logic, and backtest the configuration before I consider running it.\n\nThe AI assistant interprets the intention behind the request and uses the authorised Coinrule MCP tools required to perform the relevant operations. Coinrule continues to provide the strategy, execution, validation, portfolio, and market-connectivity infrastructure underneath the conversation.\n\nThis creates a new model for AI trading: **the trader defines the objective, the AI helps translate that objective into structured actions, and Coinrule provides the controlled automation layer through which those actions are executed.**\n\n> [!WARNING]\n> A Read + Write connection can launch strategies that trade real funds. Start with Read-only access or paper trading, review every request, and make the venue and capital amount explicit.\n\n## Connect in under a minute\n\n1. Choose a compatible client such as Claude, ChatGPT, Grok, Cursor, or VS Code.\n2. Add this remote MCP server:\n\n   ```text\n   https://cloud.coinrule.com/mcp\n   ```\n\n3. Sign in to Coinrule when your client opens the OAuth window.\n4. Choose **Read-only** or **Read + Write**, then try:\n\n   ```text\n   Show my Coinrule bots and how each one is doing.\n   ```\n\nYou need a [Coinrule Cloud account](https://cloud.coinrule.com/) and an AI client that supports remote MCP connectors with OAuth. Connect an exchange or broker to trade live, or use paper trading without one.\n\n## Agent Plugin\n\nThis repository is packaged as an [Agent Plugins 1.0.0](https://agent-plugins.org/) plugin for compatible clients and directories, including Cursor:\n\n- [`plugin.json`](plugin.json) contains portable plugin metadata.\n- [`mcp.json`](mcp.json) declares the hosted Coinrule MCP server using the Streamable HTTP transport.\n- [`server.json`](server.json) contains the official MCP Registry metadata.\n\nAgent Plugins 1.0.0 intentionally leaves OAuth to the client. When the plugin connects to `https://cloud.coinrule.com/mcp`, the client discovers Coinrule's OAuth 2.1 endpoints and asks the user to authorize Read-only or Read + Write access.\n\n---\n\n## From Trading Bots to AI Trading Agents\n\nTraditional trading automation usually begins with configuration.\n\nA trader chooses an asset, selects an indicator, defines an entry condition, configures an exit, chooses a position size, and decides where the strategy should run.\n\nThis approach can be effective, but it requires the trader to translate an investment or trading thesis into the structure expected by the software.\n\nAI trading introduces a different interaction model.\n\nInstead of beginning with a configuration screen, the process can begin with intent.\n\nFor example:\n\n> I want to investigate whether short-term weakness in Ethereum creates attractive entries when the longer-term trend remains bullish.\n\nAn AI assistant can help break this idea into questions:\n\n- How should short-term weakness be measured?\n- How should the broader trend be defined?\n- Which timeframe should be used?\n- What conditions invalidate the setup?\n- How much capital should be allocated?\n- How frequently should the strategy be allowed to trade?\n- What stop-loss or exit logic should apply?\n- How would the strategy have behaved historically?\n\nThe conversation can then move from an abstract idea toward a structured trading strategy.\n\nWith Coinrule MCP, the AI assistant can interact with authorised Coinrule functionality during this process. Depending on the permissions granted, that may include reviewing portfolio information, examining existing strategies, creating strategies, running backtests, launching baskets, or managing trading automation.\n\nThe important distinction is that the language model is not replacing the trading infrastructure.\n\nThe AI assistant acts as the conversational and reasoning interface. Coinrule remains the underlying system responsible for connecting strategies with supported exchanges, brokers, assets, portfolio data, and automation functionality.\n\n---\n\n## What Is Coinrule MCP?\n\n**MCP** stands for **Model Context Protocol**.\n\nIt is an open protocol designed to allow compatible AI applications to interact with external tools and services.\n\nCoinrule operates an official remote MCP server that allows supported AI assistants to connect to Coinrule Cloud. Once the connection has been authorised, the assistant can use the Coinrule tools available within the permission scope selected by the user.\n\nFrom the trader's perspective, the interaction remains conversational.\n\nYou might ask:\n\n> Show me my active Coinrule strategies and summarise their recent activity.\n\nThe AI assistant determines which authorised tools are required, requests the information through Coinrule MCP, and then presents the results in a conversational format.\n\nYou do not necessarily need to know the names of the underlying tools or understand Coinrule's internal API structure.\n\nThis abstraction is one of the most significant aspects of AI-assisted trading.\n\nThe trader can concentrate on **what they want to understand or accomplish**, while the AI assistant handles much of the translation between natural language and structured software operations.\n\nCoinrule MCP is therefore not a separate exchange, broker, trading account, or custody solution. It is another interface through which authorised Coinrule functionality can be accessed.\n\nStrategies created or managed through an AI assistant remain part of the broader Coinrule trading environment rather than existing in a disconnected AI-only system.\n\n---\n\n## How AI Trading Through Coinrule MCP Works\n\nA typical MCP interaction can be understood as a four-layer process.\n\n### 1. The Trader Defines an Intention\n\nYou begin by describing a task in natural language.\n\nFor example:\n\n> Analyse my current strategies and identify which ones have produced signals during the last 24 hours.\n\nOr:\n\n> Help me design a trend-following strategy for BTC using a moving-average filter and a volatility-based risk condition.\n\nThe prompt can be simple or highly detailed.\n\n### 2. The AI Assistant Interprets the Request\n\nThe language model analyses the instruction and determines what information or actions may be necessary.\n\nFor an account-analysis request, it may need portfolio or strategy information.\n\nFor strategy development, it may need to construct and validate a configuration.\n\nFor historical analysis, it may need to run or retrieve a backtest.\n\n### 3. Coinrule MCP Provides Authorised Tools\n\nThe assistant can call only the Coinrule functionality permitted by the user's connection.\n\nA Read-only connection is intended for information retrieval and analysis.\n\nA Read + Write connection provides additional capabilities for supported strategy creation and management actions.\n\n### 4. Coinrule Handles the Underlying Trading Infrastructure\n\nCoinrule remains the operational layer responsible for the relevant strategy, account, portfolio, backtesting, and market-connectivity functionality.\n\nThe AI assistant becomes an interface to that infrastructure rather than receiving unrestricted access to the trading environment.\n\nThis separation matters.\n\nAn AI model may be effective at interpreting objectives and helping users reason about strategies, but trading actions still need structured infrastructure, authentication, permissions, validation, and execution controls.\n\nCoinrule MCP is designed to connect these two layers: **AI reasoning and trading automation**.\n\n---\n\n# The AI Trading Lifecycle\n\nOne of the most useful ways to understand Coinrule MCP is to look at the complete lifecycle of an AI-assisted trading strategy.\n\nInstead of viewing AI trading as a single prompt that immediately produces a trade, a more disciplined workflow can involve multiple stages.\n\n## 1. Understand the Current Portfolio\n\nBefore creating another strategy, an AI assistant can help you understand what is already running.\n\nWith appropriate read permissions, you may be able to ask questions about balances, holdings, open positions, strategies, performance, and recent trading activity.\n\nFor example:\n\n> Summarise my current Coinrule portfolio and identify my largest asset exposures.\n\nOr:\n\n> Show my active strategies, their current status, and the most recent signal generated by each one.\n\nThis turns the AI assistant into a conversational layer for portfolio intelligence.\n\nRather than manually moving between individual accounts and strategies, the trader can begin with a broader question and progressively investigate the areas that matter.\n\nA conversation might evolve like this:\n\n> Which strategy currently has the largest open exposure?\n\nThen:\n\n> What conditions caused that strategy to enter?\n\nThen:\n\n> Show me its recent activity.\n\nThen:\n\n> Compare its behaviour with my other momentum strategies.\n\nThis ability to move from a high-level portfolio question into increasingly detailed analysis is an important characteristic of conversational AI trading.\n\n---\n\n## 2. Express the Trading Thesis in Natural Language\n\nMost trading strategies begin as ideas rather than software configurations.\n\nExamples might include:\n\n- Buy temporary weakness during a long-term uptrend.\n- Reduce exposure when market volatility rises significantly.\n- Accumulate an asset periodically rather than attempting to time every entry.\n- Follow strong momentum but exit when the trend deteriorates.\n- Rebalance a portfolio when allocations move outside predefined ranges.\n\nAn AI assistant can help convert these ideas into more precise specifications.\n\nFor example:\n\n> I want to buy significant dips in BTC, but only when the longer-term trend remains positive.\n\nThe assistant can help identify the missing variables.\n\nWhat constitutes a significant dip?\n\nWhat defines the long-term trend?\n\nHow frequently can the strategy enter?\n\nShould the strategy scale into positions?\n\nWhere should risk be limited?\n\nWhen should profits be taken?\n\nThe advantage of this approach is that strategy creation can become an iterative conversation rather than a one-time configuration exercise.\n\nThe trader can refine the strategy until its logic is explicit enough to validate and test.\n\n---\n\n## 3. Translate the Idea Into Systematic Rules\n\nAI trading becomes considerably more useful when an idea can be converted into deterministic rules.\n\nA systematic strategy might include:\n\n- Market or asset selection\n- Entry conditions\n- Exit conditions\n- Technical indicators\n- Timeframes\n- Position sizing\n- Capital allocation\n- Maximum trading frequency\n- Cooldown periods\n- Stop-loss logic\n- Take-profit logic\n- Portfolio exposure limits\n\nAn AI assistant can act as a bridge between qualitative thinking and quantitative structure.\n\nThe goal is not to let the AI produce vague instructions such as:\n\n> Buy when the market looks good.\n\nInstead, the conversation should move toward explicit logic.\n\nFor example:\n\n> Enter when RSI indicates short-term weakness, but only when price remains above a longer-term moving-average trend filter. Limit each position to a defined percentage of available capital and include an exit if the trend condition is no longer valid.\n\nThe more precise the strategy becomes, the easier it is to review, test, and monitor.\n\n---\n\n## 4. Validate the Strategy Before Execution\n\nAn AI-generated trading idea should not automatically be treated as a valid strategy.\n\nAI systems can misunderstand ambiguous requests, select inappropriate assumptions, or generate logic that appears reasonable conversationally but behaves differently when translated into actual trading rules.\n\nValidation is therefore a critical stage.\n\nBefore execution, the trader should review questions such as:\n\n- Does the resulting strategy match the original intention?\n- Are the entry conditions explicit?\n- Are the exit conditions complete?\n- Is position sizing defined?\n- Is maximum capital exposure limited?\n- Could multiple conditions trigger unexpectedly?\n- Does the strategy behave differently during high volatility?\n- What happens if the market gaps or moves rapidly?\n- What happens to existing positions when the strategy is stopped?\n\nA sophisticated AI trading workflow should use the assistant not merely to create strategies, but also to **explain them**.\n\nFor example:\n\n> Explain this strategy back to me in plain English before doing anything else.\n\nOr:\n\n> Identify the three largest risks or ambiguities in this configuration.\n\nOr:\n\n> Describe exactly what would cause an entry, what would cause an exit, and what would happen if the strategy were stopped while a position remained open.\n\nThis review step creates a human-in-the-loop process in which AI accelerates strategy development without eliminating trader oversight.\n\n---\n\n## 5. Backtest the Trading Logic\n\nOnce the strategy has been clearly defined, historical testing can provide additional information.\n\nCoinrule MCP includes functionality related to backtesting and allows compatible assistants to work with historical strategy simulations through authorised tools.\n\nA trader might ask:\n\n> Backtest this strategy and summarise the results.\n\nA more analytical request could be:\n\n> Compare several versions of this strategy using different entry thresholds and explain how the changes affect trade frequency, drawdown, and overall behaviour.\n\nThe purpose of backtesting is not to discover a configuration that appears perfect historically.\n\nExcessive optimisation can produce strategies that fit historical data extremely well but fail under future market conditions.\n\nInstead, backtesting can be used to investigate questions such as:\n\n- Did the strategy behave as expected?\n- How frequently did it trade?\n- Under what market conditions did it struggle?\n- Did small parameter changes dramatically alter the results?\n- Were losses concentrated in particular market regimes?\n- Did the risk controls behave as intended?\n\nHistorical results should be treated as evidence about past behaviour, not a guarantee of future profitability.\n\n---\n\n## 6. Consider Paper Trading or Controlled Testing\n\nBefore deploying AI-generated automation with real capital, a trader may prefer to test the workflow in a simulated environment.\n\nThis is particularly useful when learning how conversational strategy management works.\n\nA controlled process could be:\n\n1. Describe the strategy.\n2. Ask the AI assistant to formalise the rules.\n3. Review the resulting logic.\n4. Validate the configuration.\n5. Run historical tests.\n6. Test the strategy using paper trading where available.\n7. Monitor how the strategy responds to live market conditions.\n8. Review signals and activity.\n9. Refine the configuration.\n10. Consider live deployment only after independent review.\n\nThis approach treats AI as a strategy-development and automation tool rather than assuming that every AI-generated idea should immediately control real capital.\n\n---\n\n## 7. Launch and Monitor the Trading Agent\n\nOnce a strategy has been reviewed and the appropriate permissions are available, Coinrule MCP can support workflows that extend beyond strategy creation.\n\nThe AI assistant can become an ongoing interface for monitoring the trading automation.\n\nYou might ask:\n\n> Which of my strategies generated signals today?\n\nOr:\n\n> Summarise the recent activity across all my running strategies.\n\nOr:\n\n> Compare the recent performance of my momentum and mean-reversion strategies.\n\nThe assistant can help transform raw trading activity into a more understandable summary.\n\nThis is where the concept of an **AI trading agent** becomes particularly relevant.\n\nThe conversational workflow can continue across the strategy's lifecycle:\n\n**Idea → Definition → Validation → Backtest → Deployment → Monitoring → Review → Adjustment**\n\n---\n\n# What Can an AI Trading Assistant Do With Coinrule MCP?\n\nThe precise functionality available depends on the authorised connection and supported Coinrule tools, but the MCP workflow can cover several important areas.\n\nFor the complete, scope-marked list of tools exposed by the hosted server, see the [tools reference](docs/TOOLS.md).\n\n## Portfolio Intelligence\n\nA connected assistant can help retrieve and summarise supported account information.\n\nThis may allow traders to investigate:\n\n- Balances\n- Holdings\n- Portfolio exposure\n- Open positions\n- Connected accounts\n- Recent strategy activity\n- Signals\n- Performance information\n\nInstead of treating portfolio monitoring as a collection of dashboards, traders can investigate their accounts conversationally.\n\nFor example:\n\n> Give me a high-level summary of my portfolio.\n\nThen:\n\n> Which positions contribute the most exposure?\n\nThen:\n\n> Which automated strategies are responsible for those positions?\n\nThis creates a more exploratory approach to portfolio analysis.\n\n---\n\n## Strategy Analysis\n\nAn AI assistant can also act as a conversational interface for understanding existing trading automation.\n\nQuestions might include:\n\n> Which strategies are currently active?\n\n> Which strategy most recently opened a position?\n\n> Summarise the activity of my strategies over the last day.\n\n> Which strategies have not generated signals recently?\n\nThis can be particularly useful for traders operating multiple automated systems simultaneously.\n\nRather than reviewing each system independently, the AI assistant can help organise the information around the trader's question.\n\n---\n\n## AI-Assisted Strategy Creation\n\nWith the required permissions, natural-language descriptions can become the starting point for new automated strategies.\n\nFor example:\n\n> Create a conservative DCA concept that increases purchases during larger market declines but limits total portfolio exposure.\n\nOr:\n\n> Design a trend-following strategy that requires confirmation from both momentum and volume before entering.\n\nThe trader can then continue refining the strategy conversationally.\n\n> Add a cooldown period.\n\n> Reduce the maximum capital allocation.\n\n> Explain when the strategy would stop entering new positions.\n\n> Backtest the revised version.\n\nThe important advantage is iteration.\n\nStrategy development becomes less about completing a configuration form correctly on the first attempt and more about progressively improving the specification through dialogue.\n\n---\n\n## Backtest Analysis and Comparison\n\nAI assistants can also help make historical testing more conversational.\n\nInstead of simply generating a result, the assistant can be asked to interpret differences between scenarios.\n\nFor example:\n\n> Compare a conservative and aggressive version of this strategy.\n\nOr:\n\n> Test three different entry thresholds and explain how each changes the number of trades.\n\nOr:\n\n> Which parameter appears most sensitive to small changes?\n\nThe AI layer can help traders reason about backtest results rather than merely viewing them.\n\nHowever, any interpretation remains subject to the limitations of historical simulation.\n\nBacktests can be affected by market regime, data quality, parameter selection, execution assumptions, and overfitting.\n\nThey should therefore be treated as analytical tools rather than predictions.\n\n---\n\n## Strategy Templates and Starting Points\n\nNot every trading strategy needs to begin from a blank page.\n\nCoinrule's ecosystem includes predefined trading strategies and templates that can be used as starting points.\n\nAn AI-assisted workflow could therefore begin with exploration.\n\nFor example:\n\n> Find a strategy template related to dollar-cost averaging and explain its basic logic.\n\nOr:\n\n> Show me approaches that could be relevant to momentum trading, then compare their assumptions.\n\nThe trader can select a starting point and use conversation to modify it.\n\nThis can reduce the distance between discovering a strategy concept and adapting it to a specific objective.\n\n---\n\n## Multi-Asset Trading and Baskets\n\nAI trading is not limited to individual buy and sell decisions.\n\nCoinrule MCP can also support workflows involving diversified groups of assets and target allocations.\n\nA trader might describe an allocation concept in natural language and then refine it.\n\nFor example:\n\n> Create a diversified basket concept using several assets with defined target weights.\n\nThen:\n\n> Explain the concentration risk.\n\nThen:\n\n> Show how the allocation changes if I reduce the largest position.\n\nThis makes AI useful not only for individual trading signals but also for portfolio construction and systematic allocation workflows.\n\n---\n\n# AI Trading Requires Risk Controls\n\nThe phrase **AI trading agent** can imply autonomy, but responsible automation requires boundaries.\n\nAn effective trading agent should operate within clearly defined rules and permissions.\n\nThat means separating several concepts:\n\n### Analysis\n\nThe AI can inspect information and help the trader understand what is happening.\n\n### Recommendation or Strategy Design\n\nThe AI can help formulate, explain, or compare potential trading logic.\n\n### Validation\n\nThe strategy can be reviewed and tested before execution.\n\n### Execution\n\nThe system can perform authorised actions when the appropriate permissions are available.\n\nThese stages should not automatically be treated as interchangeable.\n\nA trader may be comfortable using AI for portfolio analysis without granting the ability to create or manage strategies.\n\nAnother trader may use AI for strategy creation but still manually review every configuration before deployment.\n\nThe appropriate level of automation depends on the user's objectives and risk tolerance.\n\n---\n\n## Read-Only vs Read + Write Access\n\nCoinrule MCP uses OAuth 2.1 scopes to control what a connected assistant can do.\n\n| Access level | OAuth scope | Your assistant can do |\n| --- | --- | --- |\n| **Read-only** | `coinrule:read` | Inspect strategies, balances, holdings, trades, signals, PnL, and saved backtests. It cannot place or change trades. |\n| **Read + Write** | `coinrule:read` + `coinrule:write` | Everything in Read-only, plus create, launch, update, start, and stop strategies; launch baskets; and run backtests. |\n\nWith a **Read-only** connection, the assistant can access supported information without receiving the write capabilities required to create or modify trading automation. This is often the appropriate starting point for users who primarily want conversational portfolio analysis or who are exploring MCP functionality.\n\nWith **Read + Write** access, supported management and trading-automation tools become available. This creates substantially greater capability, but also greater responsibility. When write permissions are connected to live trading infrastructure, instructions may ultimately affect real capital.\n\nUsers should therefore verify the requested action before allowing consequential changes.\n\n---\n\n# Security and the Architecture of AI Trading\n\nConnecting an AI assistant to a trading platform raises an important question:\n\n**What does the AI actually have access to?**\n\nCoinrule MCP uses an OAuth-based permission model in which the AI assistant receives authorised, scoped access rather than direct access to the user's Coinrule password or exchange API keys.\n\nConceptually, the architecture can be understood as:\n\n**Trader**\n\n↓\n\n**AI Assistant**\n\n↓\n\n**Authorised MCP Tools**\n\n↓\n\n**Coinrule Trading Infrastructure**\n\n↓\n\n**Connected Markets and Accounts**\n\nThis separation is important because an AI trading system should not rely on giving a language model unrestricted access to sensitive account credentials.\n\nInstead, the assistant operates through defined tools and permissions.\n\nUsers should still regularly review their connected applications and [revoke access](https://cloud.coinrule.com/settings/integrations) that is no longer required.\n\n---\n\n# How to Connect Coinrule MCP\n\nThe exact interface varies depending on the AI assistant, but the connection uses the same official remote server.\n\n## Before You Begin\n\nYou will generally need:\n\n- A Coinrule Cloud account\n- A compatible AI assistant or MCP client\n- Support for remote MCP connections and OAuth authentication\n- An exchange or broker connection when live execution is required\n\n## Add Coinrule to your client\n\n[![Add to Cursor](https://img.shields.io/badge/Cursor-Add_Coinrule-000000?style=for-the-badge)](cursor://anysphere.cursor-deeplink/mcp/install?name=coinrule&config=eyJ1cmwiOiJodHRwczovL2Nsb3VkLmNvaW5ydWxlLmNvbS9tY3AifQ==)\n[![Add to VS Code](https://img.shields.io/badge/VS_Code-Add_Coinrule-007ACC?style=for-the-badge)](vscode:mcp/install?%7B%22name%22%3A%22coinrule%22%2C%22type%22%3A%22http%22%2C%22url%22%3A%22https%3A%2F%2Fcloud.coinrule.com%2Fmcp%22%7D)\n\n| Client | How to connect | Help |\n| --- | --- | --- |\n| **Cursor** | Use the button above, or add the remote server from [`examples/cursor-mcp.json`](examples/cursor-mcp.json). | [Cursor configuration](https://cursor.com/docs/mcp) |\n| **VS Code** | Use the button above, or add the server to `.vscode/mcp.json` from [`examples/vscode-mcp.json`](examples/vscode-mcp.json). | [VS Code configuration](https://code.visualstudio.com/docs/agents/reference/mcp-configuration) |\n| **Claude** | In Claude's connector settings, add a remote MCP connector and paste the server URL. | [Connect Claude](https://cloud.coinrule.com/docs/mcp/connect-claude) |\n| **ChatGPT** | Go to **Settings → Apps / Connectors**, add a custom connector, and paste the server URL. | [Connect ChatGPT](https://cloud.coinrule.com/docs/mcp/connect-chatgpt) |\n| **Grok** | In a Grok client with remote MCP support, add a custom tool or remote MCP server and paste the URL. | [Connect Grok](https://cloud.coinrule.com/docs/mcp/connect-grok) |\n| **Another MCP client** | Use the client’s remote-MCP configuration and adapt [`examples/mcp-generic.json`](examples/mcp-generic.json) if it uses an `mcpServers` map. | [General connection guide](https://cloud.coinrule.com/docs/mcp/connecting-your-ai-assistant) |\n\nOAuth starts when the client first connects. Do not add a Coinrule password or exchange API key to an MCP configuration file.\n\n**ChatGPT `invalid_client`?** Remove the Coinrule connector and add it again from scratch. This refreshes the client registration. See the [ChatGPT guide](https://cloud.coinrule.com/docs/mcp/connect-chatgpt) or [troubleshooting](https://cloud.coinrule.com/docs/mcp/troubleshooting).\n\n## Connect manually\n\n1. Open the area where the AI application allows external tools, apps, connectors, integrations, or remote MCP servers to be added.\n2. Add Coinrule's official MCP server endpoint:\n\n   ```text\n   https://cloud.coinrule.com/mcp\n   ```\n\n3. Complete the Coinrule OAuth flow and choose the minimum permission level needed for your use case.\n4. Begin with a low-risk read operation, such as:\n\n   > List my Coinrule strategies.\n\nThis confirms the connection before you attempt more sophisticated workflows.\n\n---\n\n# How to Prompt an AI Trading Agent Effectively\n\nThe quality of an AI-assisted trading workflow depends heavily on the quality of the instructions.\n\nA vague request leaves important variables open to interpretation.\n\nFor example:\n\n> Build me a profitable Bitcoin strategy.\n\nThis request does not specify:\n\n- Trading timeframe\n- Capital allocation\n- Market venue\n- Risk tolerance\n- Entry conditions\n- Exit conditions\n- Position sizing\n- Maximum loss\n- Strategy frequency\n- Whether trading should be live or simulated\n\nA more useful instruction would be:\n\n> Help me design a BTC trend-following strategy for research purposes. Use a four-hour timeframe, define an objective trend filter, limit individual positions to a predefined percentage of capital, include explicit entry and exit conditions, and explain every rule before running a historical backtest. Do not launch anything live.\n\nThe second prompt establishes boundaries.\n\nThat is especially important when an assistant has write permissions.\n\n---\n\n## Useful Information to Include in AI Trading Prompts\n\nWhen relevant, specify:\n\n### Objective\n\nWhat are you trying to achieve?\n\nTrend following, accumulation, mean reversion, diversification, risk reduction, or another goal?\n\n### Asset\n\nWhich cryptocurrency, stock, ETF, or market should the strategy consider?\n\n### Venue\n\nWhich connected exchange or broker should be used?\n\n### Timeframe\n\nIs the strategy operating over minutes, hours, days, or longer periods?\n\n### Entry Logic\n\nWhat conditions must be satisfied before entering?\n\n### Exit Logic\n\nWhat conditions should close or reduce the position?\n\n### Capital Allocation\n\nHow much capital may the strategy use?\n\n### Position Sizing\n\nHow large can each individual trade become?\n\n### Risk Controls\n\nShould the strategy include a stop loss, allocation cap, cooldown period, maximum number of entries, or other limits?\n\n### Testing Requirements\n\nShould the strategy be validated, backtested, or paper traded before any live action?\n\n### Execution Permission\n\nAre you requesting analysis only, strategy creation, simulation, or actual deployment?\n\nExplicitly stating this can help reduce ambiguity.\n\n---\n\n# Example AI Trading Prompts\n\n## Portfolio Analysis\n\n> Review my Coinrule portfolio, summarise my largest exposures, list my active strategies, and highlight any recent signals. Do not make any changes.\n\n## Strategy Investigation\n\n> Explain what my currently running momentum strategy is designed to do, summarise its recent activity, and identify its main risk controls.\n\n## Strategy Design\n\n> Design a rule-based ETH momentum strategy using objective technical conditions. Include a trend filter, an entry condition, an exit condition, position-sizing limits, and a cooldown mechanism. Explain the complete logic before creating anything.\n\n## Backtest Comparison\n\n> Compare three versions of this strategy using conservative, moderate, and aggressive entry thresholds. Run historical tests and summarise differences in trade frequency and risk characteristics. Do not launch a live strategy.\n\n## Risk Review\n\n> Review this proposed strategy and identify ambiguous instructions, potential overtrading risks, concentration risks, and any conditions that could lead to unexpectedly large exposure.\n\n## Strategy Monitoring\n\n> Summarise all signals generated by my running strategies during the last 24 hours and explain which strategies were most active.\n\n## Basket Research\n\n> Help me design a diversified multi-asset basket. Explain the reasoning behind each target allocation and identify where concentration risk could arise before creating the basket.\n\nThese examples illustrate an important principle:\n\n**The best AI trading prompts describe both the objective and the boundaries.**\n\nFind more copy-paste examples in [docs/PROMPTS.md](docs/PROMPTS.md).\n\n---\n\n# Best Practices for AI-Assisted Trading\n\n## Begin With Observation\n\nBefore allowing an AI assistant to manage strategies, use it to inspect your portfolio and understand your existing automation.\n\nThis provides an opportunity to learn how the assistant interprets your requests.\n\n## Use the Minimum Necessary Permissions\n\nIf analysis is sufficient, use Read-only access.\n\nExpand permissions only when additional functionality is genuinely required.\n\n## Separate Research From Execution\n\nBe explicit about whether you are asking the AI to:\n\n- Explain an idea\n- Design a strategy\n- Validate a configuration\n- Run a backtest\n- Create a strategy\n- Launch a strategy\n\nThese are different actions.\n\n## Require Explanations\n\nBefore executing a consequential strategy, ask the assistant to explain exactly what the strategy will do.\n\n## Define Risk Before Entry\n\nRisk controls should be part of the initial strategy design rather than an afterthought.\n\n## Test Before Going Live\n\nBacktesting and paper trading can help expose problems before real capital is involved, although neither can guarantee future results.\n\n## Monitor Running Automation\n\nAn automated strategy should still be reviewed.\n\nUse the AI assistant to investigate signals, positions, performance, and unexpected activity.\n\n## Treat AI Output as Fallible\n\nLanguage models can misunderstand instructions and generate incorrect analysis.\n\nAI should therefore accelerate the trading workflow without eliminating independent judgement.\n\n---\n\n# Who Is Coinrule MCP For?\n\nCoinrule MCP may be useful for several types of traders.\n\n## Traders Managing Multiple Automated Strategies\n\nAn AI assistant can provide a single conversational interface for investigating strategy activity.\n\n## Traders Who Prefer Natural Language\n\nUsers who understand their trading objective but do not want every interaction to begin with dashboard navigation may find conversational control more intuitive.\n\n## Strategy Researchers\n\nAI can help turn an informal hypothesis into a more structured strategy specification that can then be reviewed and tested.\n\n## Systematic Traders\n\nUsers already familiar with rule-based trading can use AI to accelerate strategy iteration, comparison, monitoring, and management.\n\n## Portfolio-Focused Traders\n\nConversational portfolio queries and multi-asset basket workflows can help users reason about exposure at a broader level.\n\n## Traders Exploring Agentic AI\n\nFor users interested in AI agents, MCP creates a practical connection between language-model reasoning and authorised trading tools.\n\nThe common theme is not complete autonomy.\n\nIt is the ability to use AI as a more intelligent interface for systematic trading infrastructure.\n\n---\n\n# Coinrule MCP and the Future of Agentic Trading\n\nThe long-term significance of MCP extends beyond convenience.\n\nTrading platforms have traditionally required humans to learn the software's interface.\n\nUsers adapt their thinking to menus, forms, dashboards, and configuration structures.\n\nConversational AI reverses part of that relationship.\n\nThe trader begins with an objective.\n\nThe AI interprets the objective.\n\nExternal tools provide the capabilities required to act on it.\n\nThe result is an emerging form of **agentic trading**, where AI assistants can participate throughout the lifecycle of a systematic strategy while operating through defined permissions and trading infrastructure.\n\nA future AI trading workflow may increasingly resemble a conversation:\n\n> What is happening in my portfolio?\n\n> Which strategy is responsible?\n\n> Why did it enter?\n\n> How would the strategy behave with a tighter risk limit?\n\n> Compare the alternatives.\n\n> Backtest the revised version.\n\n> Explain the results.\n\n> Prepare the strategy for review.\n\nThe interface changes from a collection of isolated actions into a continuous reasoning process.\n\nThe opportunity is not simply to make trading faster.\n\nIt is to make the relationship between **human intent, artificial intelligence, systematic rules, and automated execution** more direct.\n\n---\n\n# Frequently Asked Questions\n\n## Is Coinrule MCP an AI Trading Bot?\n\nNot exactly.\n\nCoinrule MCP is the connection layer that allows compatible AI assistants to interact with authorised Coinrule trading functionality.\n\nThe AI assistant provides the conversational interface, while Coinrule provides the underlying strategy and automation infrastructure.\n\n## Can I Use AI to Analyse My Coinrule Account Without Allowing It to Trade?\n\nYes.\n\nA Read-only permission level allows the assistant to inspect supported account and strategy information without granting write access for strategy-management actions.\n\n## Can an AI Assistant Create Trading Strategies?\n\nWith the appropriate tools and permissions, Coinrule MCP can support strategy creation and management workflows through natural-language instructions.\n\n## Can I Backtest Strategies Through an AI Conversation?\n\nYes. Backtesting can be included in the conversational strategy-development workflow, allowing the user to move from idea generation to historical testing and comparison.\n\n## Can Coinrule MCP Work With Different AI Assistants?\n\nCoinrule MCP is designed for compatible AI clients that support remote MCP connections.\n\n## Does the AI Assistant Receive My Coinrule Password or Exchange API Keys?\n\nThe connection uses scoped authentication rather than giving the AI assistant direct access to sensitive account credentials such as your Coinrule password or exchange API keys.\n\n## Can AI-Generated Strategies Lose Money?\n\nYes.\n\nAI-generated trading strategies can contain errors, rely on incorrect assumptions, behave differently from expectations, or perform poorly when market conditions change.\n\nTrading itself involves significant risk.\n\nNeither artificial intelligence, trading automation, historical backtests, nor simulations guarantee profitable results.\n\n## Should I Let an AI Trading Agent Operate Autonomously?\n\nThat is a risk and permission decision for the individual user.\n\nA more controlled approach is to begin with Read-only access, use explicit prompts, independently review strategies, test configurations, define strict risk parameters, and understand every consequential action before permitting live execution.\n\n---\n\n# Supported venues and assets\n\nUse Coinrule MCP with the accounts you have connected to Coinrule, including:\n\n- Binance\n- Coinbase\n- Kraken (spot and perpetuals)\n- Hyperliquid\n- KuCoin\n- Trading 212\n\nSupported asset types include crypto spot, perpetuals, and stocks, with paper-trading equivalents. See [exchanges and connections](https://cloud.coinrule.com/docs/exchanges) for current venue support and setup instructions.\n\n---\n\n# Distribution\n\nThis repository contains a standards-based [`server.json`](server.json) manifest for the hosted Streamable HTTP server. It is ready to support publication in the [official MCP Registry](https://modelcontextprotocol.io/registry/about); registry publishing and namespace verification are separate release steps.\n\n---\n\n# Coinrule MCP Resources\n\n- [Coinrule MCP overview](https://coinrule.com/mcp/)\n- [Official MCP Server](https://cloud.coinrule.com/mcp)\n- [MCP documentation](https://cloud.coinrule.com/docs/mcp)\n- [Connection guide](https://cloud.coinrule.com/docs/mcp/connecting-your-ai-assistant)\n- [Tools reference](https://cloud.coinrule.com/docs/mcp/tools-reference)\n- [Example prompts](https://cloud.coinrule.com/docs/mcp/example-prompts)\n- [AI Trading Agent](https://coinrule.com/ai-trading-agent/)\n- [Pricing](https://cloud.coinrule.com/pricing)\n- [Support](https://cloud.coinrule.com/support)\n- [Coinrule homepage](https://coinrule.com)\n\n---\n\nThis repository and the Coinrule MCP are provided by [Coinrule](https://coinrule.com/), AI-powered trading automation backed by Y Combinator.\n\n> **Risk notice:** AI-generated analysis, automated trading strategies, and algorithmic systems can contain errors and may produce unexpected results. Trading cryptocurrencies, stocks, ETFs, derivatives, and other financial instruments involves risk and may result in loss of capital. Backtests, simulations, hypothetical results, and past performance do not guarantee future outcomes. Review all strategies independently, understand the permissions granted to connected AI assistants, and carefully verify consequential actions before using real funds. Nothing in this article constitutes financial, investment, or trading advice.\n",
  "bytes": 40963,
  "sha": "c1d4c169cefa71cdd688c1caee7ee2e8fbb95c6f5b118078789b92ab80d9c979",
  "repo_slug": "coinrule-com/coinrule-mcp",
  "fonte": "repo",
  "truncated": false,
  "api": "https://agentalog.com/api/listings/mcp_io_github_coinrule_com_coinrule_mcp_7c845c3c/readme"
}