Summary

Calling help(action="tools") (and help(action="tools", category=..., format="full")) returns only a count, e.g. Found 20 tools. It does not return the tool names, descriptions, parameter schemas, or usage examples.

This makes the MCP hard to consume programmatically and for agents to self-discover capabilities. An agent cannot tell which tools exist or how to call them without prior out-of-band knowledge (e.g. a local SKILL.md).

Requested improvement

Make help(action="tools") return a consumable catalog, one or more of:

1. Tool list with descriptions — name + one-line description + key parameters for each tool/action.
2. Usage examples — at least one example call per tool/action.
3. MCP Resources — expose consumable resources (per the MCP resources spec, https://modelcontextprotocol.io/specification/2025-11-25/server/resources) that document the toolset, schemas, and examples, so any MCP client can read them without bespoke parsing.

The goal: an agent (or any MCP client) should be able to discover and correctly call every tool from the help output alone, without relying on external docs.

Context

Observed while verifying the #70 changelog fix: help(action="version") works and returns the canonical releases URL, but help(action="tools") gave no catalog — only Found 20 tools. This blocked confirming whether a richer version/changelog tool exists server-side.

Environment

• Hosted MCP server: v0.5.40
• Client: VS Code + GitHub Copilot Chat (remote HTTP MCP at https://mcp.contextstream.io/mcp)
• Related: issue Popup 'View all releases' links to github.com/contextstream/mcp/releases (404); propose release_notes tool for agents #70 (changelog/release-notes access), issue Recap Bot stalls via VS Code Copilot MCP path; recaps are session-boundary triggered, not a daily cron #72 (recap retrieval API).