{
  "markdown": "<div align=\"center\">\n\n<img src=\"assets/agentforge-mark.svg\" alt=\"AgentForge\" width=\"76\" />\n\n# agentforge-mcp\n\n**Stop re-explaining what you want to your AI coding agent.**<br/>\nHand it one rough sentence — get back a structured, tool-tuned prompt\nit can execute in one pass.\n\n[![npm](https://img.shields.io/npm/v/agentforge-mcp)](https://www.npmjs.com/package/agentforge-mcp)\n[![license](https://img.shields.io/npm/l/agentforge-mcp)](./LICENSE)\n\n[**agentforge.sciscale.org**](https://agentforge.sciscale.org) · [Quick start](#quick-start) · [npm](https://www.npmjs.com/package/agentforge-mcp)\n\n</div>\n\n![AgentForge MCP in action](assets/demo.gif)\n\n`agentforge-mcp` is the [Model Context Protocol](https://modelcontextprotocol.io)\nserver for [AgentForge](https://agentforge.sciscale.org). One tool, one job: turn\na vague request into a sharp one.\n\n## Why\n\nYou know the loop: you ask your coding agent for something, it misreads the half\nyou didn't spell out, you correct it, it breaks something else, you re-explain.\nThe fix isn't a smarter agent — it's a sharper *prompt*.\n\nAgentForge does the prompt engineering for you. Give it\n`\"add a dark mode toggle that persists\"` and it:\n\n- **extracts the real requirements** — the edge cases, the acceptance criteria,\n  the things you'd have forgotten to mention;\n- **formats for your specific agent** — Claude Code wants file:line references,\n  Cursor wants directives, Aider wants tight context;\n- **quality-checks before returning** — every prompt is scored across 12\n  dimensions before it leaves the engine.\n\nHarness engineering is having a moment — but prompt engineering matters *more*\nin the coding-agent era, not less. Tested across **1,000+ real coding cases**,\nprojects with AgentForge in the loop reached a deployable state **147% faster**.\n\n## Quick start\n\nNo install — it runs through `npx`.\n\n**1. Get an API key.** Sign in at\n[agentforge.sciscale.org](https://agentforge.sciscale.org), open **API keys**,\nand create one. The key is shown once — copy it.\n\n**2. Add it to your agent.**\n\nClaude Code:\n\n```sh\nclaude mcp add agentforge --env AGENTFORGE_API_KEY=af_your_key -- npx -y agentforge-mcp\n```\n\nCursor / Windsurf / Claude Desktop — add to your MCP config\n(`~/.cursor/mcp.json`, `~/.codeium/windsurf/mcp_config.json`, etc.):\n\n```json\n{\n  \"mcpServers\": {\n    \"agentforge\": {\n      \"command\": \"npx\",\n      \"args\": [\"-y\", \"agentforge-mcp\"],\n      \"env\": { \"AGENTFORGE_API_KEY\": \"af_your_key\" }\n    }\n  }\n}\n```\n\n**3. Use it.** Ask your agent naturally — *\"refine this with AgentForge, then\nbuild it: add a dark mode toggle that persists.\"*\n\n## The tool — `agentforge_refine_prompt`\n\n| Argument | Default | |\n|---|---|---|\n| `request` | — | Your task in plain language (1–4000 chars). Rough is fine. |\n| `target_tool` | `claude-code` | `claude-code`, `codex`, `cursor`, `aider`, `continue`, `windsurf`, `kimi`, `generic` |\n| `style` | `plan-first` | `plan-first`, `direct-edit`, `explore-first` |\n\nReturns the refined prompt, plus its Quality Engine score and your remaining\ndaily usage.\n\n## Free vs Pro\n\n| | Free | Pro |\n|---|---|---|\n| Refinements | 3 / day | Unlimited |\n| Quality Engine | scored across 12 dimensions | scored **+ auto-refined until it passes ≥ 90** |\n\n## Or use the web app\n\nDon't want to wire up an MCP server at all? The same engine — same Quality\nEngine, same per-account Pro — runs right in your browser at\n**[agentforge.sciscale.org](https://agentforge.sciscale.org)**. No install, no\nkey, no config.\n\n[![The AgentForge web app](assets/web-landing.png)](https://agentforge.sciscale.org)\n\n<div align=\"center\">\n\n### [→ Try it now at agentforge.sciscale.org](https://agentforge.sciscale.org)\n\n</div>\n\nPaste a request, pick your target tool, copy the prompt — generation history and\nadvanced modes included:\n\n![The web app in action](assets/demo-web.gif)\n\n## How it works\n\n`agentforge-mcp` is a thin client — no engine logic ships in this package. Your\nrequest goes to the AgentForge API, the hosted engine does the extraction,\nformatting, and quality-checking, and the prompt comes back. The engine keeps\nimproving without you ever updating this package.\n\n## Configuration\n\n| Variable | Default | |\n|---|---|---|\n| `AGENTFORGE_API_KEY` | — | **Required.** Your API key. |\n| `AGENTFORGE_API_URL` | `https://agentforge.sciscale.org/api/v1/refine` | Override the endpoint (rarely needed). |\n\n## Develop\n\n```sh\nnpm install\nnpm run build      # tsc -> dist/\nnode dist/index.js # runs on stdio\n```\n\n## License\n\nMIT — see [LICENSE](./LICENSE).\n\n<br/>\n\n<div align=\"center\">\n\n<a href=\"https://wow.sciscale.org\">\n  <picture>\n    <source media=\"(prefers-color-scheme: dark)\" srcset=\"assets/sciscale-white.png\" />\n    <img src=\"assets/sciscale-ink.png\" alt=\"SciScale studio\" height=\"30\" />\n  </picture>\n</a>\n\n<sub>**agentforge-mcp** is part of [AgentForge](https://agentforge.sciscale.org) — a [SciScale studio](https://wow.sciscale.org) product.</sub>\n\n</div>\n",
  "bytes": 4926,
  "sha": "98232f5ce02845300e642715313d78f386d0f07d74f476cc9f24626818f6d02a",
  "repo_slug": "sciscale-studio/agentforge-mcp",
  "fonte": "repo",
  "truncated": false,
  "api": "https://agentalog.com/api/listings/mcp_org_sciscale_agentforge_mcp_66e53a36/readme"
}