内容
# mcp-tool-catalog
**Generate human-readable docs for any MCP server's tools.**
An MCP server exposes its tools programmatically through `tools/list`, but there is
rarely a clean, human-friendly document describing what those tools do and what
arguments they take. `mcp-tool-catalog` launches a server, performs the MCP
handshake, reads its tool surface, and renders a tidy **Markdown** or **HTML**
catalog: one section per tool, with a parameter table and descriptions.
Great for READMEs, onboarding, PR reviews, and keeping a living record of what an
agent can actually do.
Part of the AAIF (Agentic AI Innovation Foundation) open-source toolset for the
agentic AI ecosystem.
## Highlights
- Works with any stdio MCP server, not a specific implementation.
- Markdown output with a table of contents and per-tool parameter tables.
- Self-contained HTML output for sharing.
- Standard library only. No pip dependencies, no markdown engine.
## Install
```bash
pip install -e .
```
Requires Python 3.9 or newer.
## Usage
Pass the command that launches your server after `--`:
```bash
mcp-tool-catalog -- mcp-mock-server --config examples/weather.json
```
Write HTML to a file:
```bash
mcp-tool-catalog --format html -o catalog.html -- mcp-mock-server
```
Write Markdown to a file:
```bash
mcp-tool-catalog -o TOOLS.md -- python my_server.py
```
## Example output
For a server exposing a `get_forecast` tool, the Markdown looks like:
```markdown
# weather-mock — tool catalog
**Server version:** 1.0.0
**Protocol:** 2025-06-18
**Tools:** 1
## `get_forecast`
Return a canned weather forecast for a city.
| Parameter | Type | Required | Description |
|-----------|------|----------|-------------|
| `city` | string | yes | |
```
## How it works
`mcp-tool-catalog` includes a minimal stdio JSON-RPC client. It sends
`initialize`, the `notifications/initialized` acknowledgement, then `tools/list`,
and renders the returned tool definitions. It never calls tools, so it is safe to
run against servers with side effects.
## Development
```bash
pip install -e .
pytest
```
Tip: `mcp-mock-server` (a companion project) is a convenient target for trying this
out without a real backend.
## License
MIT. See [LICENSE](LICENSE).
MCP 市场