Skip to main content

AI editor setup

Give your AI coding assistant knowledge of the OneSource documentation so it can answer questions and generate accurate calls without you copy-pasting from these pages.

Reading about vs. calling the API

This wires in the docs (the assistant can read about OneSource). To let the assistant call the API at runtime, install @one-source/mcp: the two complement each other, so install both for the best experience.

There are three ways to do it, in rough order of preference.

onesource-docs-mcp gives your assistant 8 structured tools, search docs, list endpoints, look up parameters, pricing, and auth, which retrieve far more precisely than plain text search. Two ways to run it:

  • Hosted, zero install: point any HTTP-capable MCP client at https://docs.onesource.io/api/mcp.
  • Local: npx -y onesource-docs-mcp (Node).

Per-editor configuration (Claude Code, Claude Desktop, Cursor, Windsurf, VS Code) lives on the Docs MCP page; start there.

2. Already using mcpdoc?​

If you've standardized on LangChain's mcpdoc across your documentation sources, add OneSource as one more entry rather than running a second kind of server. It serves keyword search over llms.txt (no structured tools) and needs Python's uv: curl -LsSf https://astral.sh/uv/install.sh | sh.

Claude Code:

claude mcp add-json onesource-docs '{"type":"stdio","command":"uvx","args":["--from","mcpdoc","mcpdoc","--urls","OneSourceDocs:https://docs.onesource.io/llms.txt","--follow-redirects"]}'

Cursor / Windsurf / Claude Desktop: add to your editor's MCP config file (VS Code uses a servers key instead of mcpServers):

{
  "mcpServers": {
    "onesource-docs": {
      "command": "uvx",
      "args": ["--from", "mcpdoc", "mcpdoc", "--urls", "OneSourceDocs:https://docs.onesource.io/llms.txt", "--follow-redirects"]
    }
  }
}

3. No MCP support: use the raw files​

If your tool can't speak MCP, point it at (or paste in) the doc bundles directly:

  • llms.txt: index of every page (lightweight, for discovery).
  • llms-full.txt: the full docs concatenated into one file (for context-window ingestion).