SuperAPI AI Directory

agentdocs-mcp

Read, create, update, comment on and share AgentDocs Markdown docs, with full-text and semantic search.

Open website

About this tool

AgentDocs MCP is the MCP server for the AgentDocs (agentdocs.eu) collaborative documentation platform, designed to make AI agents first-class citizens in document collaboration. It provides native tools for MCP clients running local servers (such as Claude Code, Claude Desktop, Cursor, Windsurf, Zed, etc.) to read, search, create, update, and share AgentDocs pages.

Main capabilities:

  • Provides 19 tools, supporting reading, searching, creating, updating, commenting on, and sharing documents.
  • Supports full-text search and semantic search.
  • Pages, spaces, and workspaces can be addressed by UUID or human-readable slug paths (such as "my-workspace/my-space/my-page").
  • Every page update creates a version on the server, and old versions can be restored from the AgentDocs UI.
  • Supports account tokens (full access) and space tokens (edit permission limited to one space, which the server automatically detects and scopes).

Usage:

  1. Remote (hosted) endpoint: no package installation is required, and any client that supports remote MCP can use it directly https://agentdocs.eu/mcp。Claude.ai(网页)、Claude Desktop and Claude mobile can add this URL via "Settings → Connectors → Add custom connector", with no token required. AgentDocs implements OAuth 2.1, and the browser will automatically open the consent page; after approval, the connection is established. Authorization covers only your documents and can be revoked at any time in agentdocs.eu → Settings → Connected apps.
  2. Local stdio server: requires an AgentDocs API token. Account tokens are obtained from agentdocs.eu → Profile → Regenerate API token; space tokens are obtained from Space settings → Tokens. Configuration examples (Claude Code, Claude Desktop, Cursor, Windsurf, Gemini CLI, VS Code Copilot, opencode, etc.) can be found in the repository documentation.
  3. For clients that do not support MCP (such as Base pi), you can use the Skill or the pure REST API.

Use cases:

  • When AI agents need to read, write, and search collaborative documents.
  • When teams use the AgentDocs platform for document collaboration and want AI to participate directly.
  • When document operations need to be integrated into MCP client workflows.

Limitations and notes:

  • The hosted endpoint rejects the path parameter because it may cause arbitrary file reads; use source_url or data instead.
  • The hosted instance may take about 15 seconds to respond to the first request after being idle (database cold start); the server has a 35-second timeout and one retry.
  • Free-tier API limits are presented with clear messages and an upgrade link.
  • The test space will be written to (creating and deleting pages), so using a temporary space is recommended.
  • Some MCP gateways (such as the Docker MCP gateway) only run servers in their catalog, and agentdocs-mcp is not yet listed, so you can use the hosted remote endpoint.
  • On Windows, some clients cannot start npx directly and need to be wrapped with cmd /c.
  • Please report security vulnerabilities privately to contact@agentdocs.eu.