Skip to content
Clawdemy

References: Tools other agents can trust

Source material

Section titled “Source material”

This lesson’s opening line about tool effectiveness and its guidance on error responses are quoted verbatim from Anthropic’s public engineering essay on writing tools; the MCP capability definitions are quoted verbatim from the official build-server tutorial. Everything else, including the bike-shop inventory server and its tools, is Clawdemy’s own synthesis and examples.

  • Writing effective tools for agents, Anthropic engineering blog, September 11, 2025. The essay behind this lesson’s description, naming, and error guidance. It opens with the line quoted in the lesson, “Agents are only as effective as the tools we give them,” advises writing descriptions as you would for a new hire, recommends prefix namespacing for large toolsets, warns that oversized or overlapping toolsets pull agents away from efficient strategies, and calls for error responses that “clearly communicate specific and actionable improvements.”
  • Build an MCP server, Model Context Protocol documentation. The official tutorial this lesson’s server section follows: the three capability kinds, with resources defined as “File-like data that can be read by clients” and tools as “Functions that can be called by the LLM (with user approval),” plus the Python walkthrough where a decorated function’s docstring becomes the tool definition.
  • Connect Claude Code to tools via MCP, Claude Code documentation. The reference for the configuration section: the local, project, and user scopes, the shared project file (.mcp.json) designed for version control, per-person approval of project-scoped servers, and environment-variable expansion syntax and locations.

Going deeper

Section titled “Going deeper”

On this site

Section titled “On this site”
  • Model Context Protocol (Track 22). The consumer side this lesson presumes: what MCP is, how clients connect to servers, and how to use the tools a server exposes.
  • Building with Claude (Track 22). Tool use from the calling side: how a model requests a tool call and how results return to the loop.

Adjacent topics

Section titled “Adjacent topics”
  • Lesson 3 of this track built the schema discipline this lesson reuses in every tool contract; if strict inputs and honest outputs are still hazy, read it before building a server.
  • The next lesson takes these tools to team scale: orchestration, where your contracts become agreements between agents and failure handling becomes the whole game.