Docs MCP endpoint
The documentation site exposes an MCP server at:MCP endpoint
Direct URL (streamable HTTP)
Clients that accept an MCP server URL directly:MCP server URL
Command-based (stdio via mcp-remote)
Clients that require a command and arguments:Setup by client
Claude Code
Claude Code
Claude Desktop
Claude Desktop
- Open Settings then Connectors.
- Click Add custom connector.
- Set the name to “WebMCP Docs” and the URL to
https://docs.mcp-b.ai/mcp. - Click Save, then Connect.
Cursor
Cursor
Add to Restart Cursor after saving.
.cursor/mcp.json in your project root:Windsurf
Windsurf
Add to Restart Windsurf after saving.
mcp_config.json:ChatGPT
ChatGPT
MCP support in ChatGPT requires a paid plan with Developer Mode enabled.
- Enable Developer Mode in ChatGPT settings.
- Go to Settings, then Connectors.
- Click Add MCP server.
- Enter
https://docs.mcp-b.ai/mcpas the server URL. - Name it “WebMCP Docs” and click Save.
Codex
Codex
Add to Restart Codex after saving.
~/.codex/config.toml:What the server provides
The MCP server gives connected agents read access to the full documentation site. Agents can search for pages, retrieve page content, and browse the site structure.llms.txt
The documentation site serves two files that follow the llmstxt.org convention for AI-readable site indexing.Files
How they are maintained
The repository ownsapps/documentation-website/llms.txt. Mintlify serves that
root-level file instead of its generated short index. Keep it concise and update
it when canonical sources or primary routes change. See Mintlify’s
custom file behavior.
Mintlify generates llms-full.txt from the published pages on each deployment.
It contains each page’s Markdown export, separated by page headers.
The maintained llms.txt includes:
- The proposal status and
document.modelContextsurface - Canonical Community Group, Chrome, OpenAI, and WPT sources
- Entry points for tutorials, how-to guides, explanations, and package references
- Links to the full Markdown export, MCP endpoint, and source repository
Usage
AI assistants (Claude, ChatGPT, Gemini, Perplexity, and others) fetchllms.txt to understand what the site covers before retrieving specific pages. If your agent supports URL-based context, point it at https://docs.mcp-b.ai/llms.txt.
The MCP endpoint and llms.txt serve different purposes. llms.txt is a static file that agents fetch over plain HTTP. The MCP endpoint is an interactive server that agents connect to as an MCP client, allowing structured queries against the documentation.
Versioning and changelog
WebMCP uses Changesets for versioning. The browser packages share one fixed release version. Changesets bumps the full group when any member changes. Private apps, templates, and test packages are not published.Where to find changelogs
Versioning policy
All packages follow Semantic Versioning:Package scope
The fixed release train contains these browser packages:
See the package index for package-specific documentation.
Reporting issues
If you find a bug or regression after upgrading:- Check existing issues.
- Open a new issue with the package name, version numbers, steps to reproduce, and expected versus actual behavior.
