Related: https://tiangolo.com/ideas/library-agent-skills/

Hi @YuriiMotov I checked out the skills folder mentioned in the above page: https://github.com/fastapi/fastapi/tree/master/fastapi/.agents/skills/fastapi. I felt it's great for someone starting out with fastapi but missing import concepts important for building production grade applications for eg: websockets, middleware etc, which are part of the official user guide.

Our we planning to either include these concepts directly in the skills or include a section which allows the agent to fetch the details about sections that our missing in the skills but present in the official user guide

Great question — this is exactly the problem that Context7 (an MCP server) was built to solve.

Context7 — live documentation fetching for AI agents

Context7 is a free, open-source MCP server that resolves a library ID and fetches its current, up-to-date documentation at query time. Because it fetches from the actual docs rather than relying on training data, the knowledge-cutoff problem disappears.

How to configure it in Claude Code

Add it to your ~/.claude/settings.json (or project-level .claude/settings.json):

{
  "mcpServers": {
    "context7": {
      "command": "npx",
      "args": ["-y", "@upstash/context7-mcp@latest"]
    }
  }
}

Then, when chatting with Claude Code, prefix your question with "use context7" to trigger a live docs fetch:

use context7 — how do I configure dependency injection with yield in FastAPI?

Claude will resolve the FastAPI library ID, fetch the relevant docs sections, and answer based on the current FastAPI documentation — not its training snapshot.

FastAPI skills (for Claude Code)

As @YuriiMotov noted, FastAPI also ships its own official agent skill at fastapi/.agents/skills/fastapi. This is complementary to Context7:

• Skills encode best practices and opinionated patterns (great for "how should I structure this?")
• Context7 fetches live reference docs (great for "what are the exact parameters for X?")

Using both together gives you pattern guidance and up-to-date API reference.

Codex CLI

Context7 works with Codex CLI too. Add it to ~/.codex/config.toml:

[[mcp_servers]]
name = "context7"
command = "npx"
args = ["-y", "@upstash/context7-mcp@latest"]

Then use the context7 tool from within a Codex session.

Other useful MCP servers for FastAPI development

• FastAPI MCP — automatically exposes your FastAPI endpoints as MCP tools (so Claude can call your own API)
• Microsoft Learn MCP — for Azure-related docs if you deploy FastAPI to Azure

The Context7 approach directly solves the manual link-pasting workflow you described.