Back to the catalog

credible

Credible analysis tools (Malloy semantic model search, query execution, and analysis skills) for your Credible workspaces.

Open source Repository Open in the app JSON README (API)

About

Credible analysis tools (Malloy semantic model search, query execution, and analysis skills) for your Credible workspaces.

Details

Kind
Plugins
Topic
No topic detected
Publisher
credibledata
Origin
marketplace
Category
ferramentas
Open pull requests
1
Last push
2026-08-26T23:18:44Z
Repository state
ativo
License
NOASSERTION
Added
2026-08-30 01:48:58
Updated
2026-08-30 01:48:58
Origin id
credibledata/credible-plugin/credible

README

# Credible

The Credible plugin connects Claude to your [Credible](https://credibledata.com) workspaces so you can ask questions about your data in natural language, get grounded answers from your team's published Malloy semantic models, and run real queries against your warehouse — without leaving the conversation.

The plugin bundles:

- A Streamable HTTP **MCP server** at `https://mcp.credibledata.com/global` exposing the core analysis tools (`list_workspaces`, `get_context`, `execute_query`, `search_malloy_docs`).
- A set of **Skills** that teach Claude how to use those tools well — phrase detection, Malloy query patterns, chart selection, analysis pitfalls, Malloy docs lookup, and the canonical analysis workflow.

## What it does

Once installed, you can ask questions like *"how many active subscribers did we have last month?"* or *"break down revenue by region for Q1"*. The plugin gives Claude the tools and skills to:

1. Discover which Credible workspaces you have access to (`list_workspaces`).
2. Search the relevant semantic model for the entities (dimensions, measures, sources) that answer the question (`get_context`).
3. Compose a valid Malloy query against that model and run it against your warehouse (`execute_query`).
4. Return rows, summary stats, and chart-ready output to the chat.

You stay in control: every query is shown before it runs, results are streamed back into the chat, and the underlying Credible permissions (org / workspace / package) are enforced server-side.

## Requirements

- An active [Credible](https://credibledata.com) account with access to at least one published workspace.
- A Claude surface that supports plugins:
  - **Claude Cowork** (Desktop)
  - **Claude Code** (Desktop UI or CLI)
  - **Claude Desktop** (chat — via "Upload plugin")
- Outbound HTTPS access to `mcp.credibledata.com`.

The first time you use the plugin, Claude will prompt you to sign in to Credible via OAuth. To revoke access, uninstall the plugin in your Claude client (Cowork → *Plugins*, Claude Desktop → *Settings → Plugins*, or `claude plugin uninstall credible@<marketplace>`).

## Install

### From the community plugin marketplace (recommended)

In **Claude Code**:

```bash
/plugin marketplace add anthropics/claude-plugins-community
/plugin install credible@claude-community
```

In **Cowork**: open *Organization settings → Plugins → Browse marketplace* and pick **Credible**.

### Manually from this repo

```bash
# Claude Code (local clone install)
git clone https://github.com/credibledata/credible-plugin.git ~/credible-plugins/credible
claude --plugin-dir ~/credible-plugins/credible

# Claude Code CLI marketplace add (no clone)
/plugin marketplace add credibledata/credible-plugin
/plugin install credible@credible
```

### From a release zip

Grab `credible.zip` from the [latest release](https://github.com/credibledata/credible-plugin/releases/latest) and upload it via:

- **Cowork** → *Organization settings → Plugins → Upload plugin*
- **Claude Desktop** → *Settings → Plugins → Upload plugin*

### OpenCode

The same Credible MCP server and analysis skills are available for
[OpenCode](https://opencode.ai). See [`opencode/README.md`](./opencode/README.md)
in this repository for installation and OAuth instructions.

## Usage examples

> Try these prompts after installing. They assume you have at least one published workspace; Claude will help you pick one if you have several.

### 1. Sanity check — explore what's available

> *"What Credible workspaces do I have access to, and what kind of data is in each?"*

Claude will call `list_workspaces` and summarize the orgs / workspaces it sees, then offer to dive into one. Good first-run prompt to confirm the OAuth flow worked end-to-end.

### 2. A natural-language data question

> *"What was our total revenue for each region in Q1 of last year?"*

Claude will use the plugin's tools and skills to identify the right `revenue` measure and `region` dimension in the workspace's semantic model (`get_context`), compose a valid Malloy query with the appropriate time filter, run it (`execute_query`), and return both the table and a renderable chart spec. The `malloy-analysis` and `malloy-charts` skills guide it through each step.

### 3. A more advanced analytical question

> *"Compute year-over-year growth in monthly active users, broken down by plan tier, for the last four quarters."*

Year-over-year comparisons aren't a one-liner in Malloy. Claude will use `get_context` to identify the right active-user measure and the `plan_tier` dimension, look up the time-shifting pattern via `search_malloy_docs` if it's unsure, then compose and run a single query via `execute_query`. The `malloy-analysis-pitfalls` skill flags caveats — for example, the ambiguity in "monthly" (calendar month vs. trailing 30 days) and partial-period under-counts.

## Tools

| Tool | What it does | Read-only |
|---|---|---|
| `list_workspaces` | Lists every Credible workspace the caller can access, grouped by org. | Yes |
| `get_context` | Searches a workspace's semantic model for sources, dimensions, measures, views, and dimensional values matching the user's question. | Yes |
| `execute_query` | Executes a Malloy query (or a predefined view) against the workspace and returns rows + render metadata. | Yes (queries cannot mutate data) |
| `search_malloy_docs` | Searches the Malloy documentation for syntax, language features, and code examples — used by Claude to look up unfamiliar patterns before writing a query. | Yes |

All tools are annotated with `readOnlyHint: true` — the plugin cannot write back to your warehouse or your Credible workspace.

## Privacy

See [PRIVACY.md](./PRIVACY.md) for the full policy. The plugin only shares the OAuth identity, the questions you ask, and the resulting query results with Credible's servers; it does not transmit credentials or warehouse data to any third party other than the LLM provider already mediating the chat. Credible's privacy policy is also published at <https://credibledata.com/privacy>.

## Support

- **Docs**: <https://docs.credibledata.com>
- **Email**: support@credibledata.com
- **Issues** (this plugin): <https://github.com/credibledata/credible-plugin/issues>

## License

See [LICENSE](./LICENSE).

More