Docs MCP gives agents read-only tools over the official Tradeics developer hub — search, fetch pages, and retrieve snippets — without touching customer workspace data.
Pair with the Tradeics Docs skill and /llms.txt.
Connection
| Field | Value (placeholder) |
|---|---|
| Transport | Streamable HTTP |
| URL | https://mcp.tradeics.example/docs |
| Auth | Authorization: Bearer <token> with docs:read (or equivalent) |
Tool surface (structure)
Populate tool names at GA; keep these capability buckets:
| Capability | Example tool id (placeholder) | Description |
|---|---|---|
| Search | docs_search |
Full-text / semantic search across guides + references |
| Fetch page | docs_get |
Fetch a path or URL from the developer hub |
| Index | docs_llms_index |
Return llms.txt structure for navigation |
| Cite | docs_resolve |
Resolve a topic → canonical path |
All tools are read-only. No mutations of workspace data.
When to use
- Agent answers “how do I…?” from official docs
- Codegen that must cite Authentication / Pagination / REST groups
- Evaluating whether a host prompt invents endpoints
When not to use
- Live RFX / catalog operations → Platform MCP
- Batch inference → LLM API
Example prompts
- “Using Docs MCP, find versioning guidance and quote the path.”
- “Resolve ‘webhook retries’ to the correct Getting Started page.”
Page sections to fill at GA
- Exact tool schemas (JSON)
- Rate limits specific to Docs MCP
- Supported locales / version pins
- Changelog for tool renames