Skip to main content
The Guild CLI includes an MCP (Model Context Protocol) server that exposes Guild operations as tools for MCP clients. Client agents call these tools to read and act on Guild resources. This page describes what each category of tool returns, so you can parse the results reliably.

​
Output format

Every tool returns its result in the content[0].text payload. The format depends on what the tool does:
  • Data-returning tools return a structured JSON string. The tool serializes the response data with JSON.stringify(data, null, 2), so the text is valid, indented JSON that a client agent can parse directly.
  • Action-taking tools return narrative plain text that describes the result of the action.
  • Errors return a plain text error string and set isError: true.
Data-returning tools return JSON, not human-readable narrative text. Parse content[0].text as JSON instead of scraping formatted output.

​
Data-returning tools

These tools return structured JSON. They cover the current user, list endpoints, and single-object get endpoints.

​
Action-taking tools

These tools perform an action and return narrative plain text describing the result.
  • guild_chat
  • guild_send_message
  • guild_interrupt_session
  • guild_add_workspace_agent
  • guild_remove_workspace_agent
  • guild_publish_context
  • guild_get_agent_readme
  • guild_get_agent_code
Prefer the terminal? Use guild api, the authenticated CLI escape hatch that calls the Guild REST API directly and prints the raw JSON response.