FixYourDocs

@fixyourdocs/mcp-server

Model Context Protocol (MCP) server that exposes a single tool — file_doc_feedback — so any MCP client (Claude Desktop, Cursor, Codex, Gemini CLI, …) can file a structured documentation-feedback report to the FixYourDocs hub at hub.fixyourdocs.io when it encounters docs that are broken, incorrect, outdated, missing, unclear, or otherwise unhelpful.

Reports are forwarded to a GitHub Issue on the doc owner's repository when they have a FixYourDocs integration installed. No phone-home, no telemetry — the server only contacts the hub when an agent explicitly calls the tool.

• Docs: docs.fixyourdocs.io/mcp/
• Spec: Docs Feedback Protocol v0
• SDK: @fixyourdocs/sdk
• Issues / contributions: https://github.com/fixyourdocs/fixyourdocs

Install

The server is designed to be run via npx, so you typically don't install it directly — your MCP client launches it on demand:

npx -y @fixyourdocs/mcp-server

It speaks the standard MCP stdio transport. There is nothing to configure unless you want to point at a different hub (see Environment variables below).

Client configuration

Each MCP client has its own config file, but the shape is the same: a named server entry with command + args.

Claude Desktop

Edit ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows):

{  "mcpServers": {    "fixyourdocs": {      "command": "npx",      "args": ["-y", "@fixyourdocs/mcp-server"]    }  }}

Restart Claude Desktop. The file_doc_feedback tool should appear in the tool list.

Cursor

Edit ~/.cursor/mcp.json (or your workspace's .cursor/mcp.json):

{  "mcpServers": {    "fixyourdocs": {      "command": "npx",      "args": ["-y", "@fixyourdocs/mcp-server"]    }  }}

Codex / OpenAI Codex CLI

Edit ~/.codex/config.toml:

[mcp_servers.fixyourdocs]command = "npx"args = ["-y", "@fixyourdocs/mcp-server"]

mcp-inspector (for verification)

npx @modelcontextprotocol/inspector npx -y @fixyourdocs/mcp-server

The inspector UI lists file_doc_feedback, accepts a fixture report, and shows the { id } returned by the hub.

Tool: file_doc_feedback

Input

Mirrors the v0 spec's Report envelope:

Output

{  "id": "01ABCDEF0123456789ABCDEFGH"}

id is a ULID returned by the hub — an idempotency confirmation, not a fetchable handle. Reports are write-only on the public API: there is no retrieval endpoint, and report contents reach the doc owner only via the hub's GitHub-issue forwarding.

Behaviour

• The server validates the input against the v0 lexical constraints (URL parsing, agent-name pattern, summary length) before contacting the hub. Validation errors are returned as MCP isError: true results with a human-readable explanation.
• On a 201 Created (new submission) or 200 OK (idempotent duplicate), the tool returns { id } regardless — agents don't need to distinguish, the hub will not double-file the issue.
• On 429 rate_limited, 5xx, or any network error, the tool surfaces the error verbatim. Clients should treat this as transient and retry with the same payload (the hub deduplicates on body content).

Environment variables

Privacy

• No telemetry, no phone-home. The server makes outbound network calls only when an agent explicitly invokes the tool. There is no analytics call, no version-check ping, no diagnostic upload.
• The report payload contains only the fields the agent supplies (doc URL, agent identity, kind, summary, optional details and suggested fix). No process environment, file paths, or system identifiers are appended.

License

Apache-2.0. See LICENSE.