#Connect FPF Reference MCP

Use this page when you want ChatGPT, Claude, an editor, or a coding CLI to retrieve bounded FPF context through the hosted MCP server.

#Hosted endpoint

https://mcp.fpf.sh/api/mcp/fpf_reference/mcp

This endpoint is the direct Vercel-hosted MCP origin over streamable HTTP. It exposes the public FPF Reference tools for catalog browsing, search, compact answers, exact generated doc reads, and index health.

The legacy endpoint is blocked during the May 2026 cost incident mitigation:

https://mcp.fpf.sh/api/mcp/fpf_memory/mcp

Keep the route entry until the scheduled compatibility review on 2026-06-30 so old clients fail cheaply with a migration signal instead of falling through to an expensive timeout. Existing users with the old URL or fpf_memory client name should move to the canonical endpoint above.

It is a JSON-RPC endpoint, not a web page. A bare browser GET returns 405 Method Not Allowed because standalone MCP SSE streams are disabled on the hosted endpoint. Paste the canonical URL into your client's MCP config; do not open it in a tab.

For a browser-readable health check, use the freshness page instead:

https://mcp.fpf.sh/api/fpf/status

It returns JSON the browser will render — source hash, snapshot age, and the upstream ref the index was built from. runtime.snapshotConsistent is an internal deployed-artifact check; upstream/currentness stays unknown there unless a monitor compares the status payload to the intended upstream ref and source hash.

Public tools:

• browse_fpf_catalog
• search_fpf
• ask_fpf
• query_fpf_spec
• read_fpf_doc
• get_fpf_index_status

#Test first

After adding the server, ask your client to call get_fpf_index_status. Then run a compact route query:

Use only fpf_reference. Call query_fpf_spec with question: "Project kickoff: align a project information system with roles and adoption next steps" and mode "compact". Return the route ID, ordered IDs, acceptance check, and next move.

A good response should include route:project-alignment in ids, then bounded next steps rather than a full FPF paste.

#ChatGPT

Use this path for ChatGPT custom apps/connectors.

1. Open ChatGPT settings.
2. Go to Apps & Connectors.
3. Enable developer mode under Advanced settings if your plan or workspace requires it.
4. Create a new custom app or connector.
5. Set the connector URL to the hosted endpoint above.
6. Create the connector, confirm the advertised tools, then open a new chat and add the connector from the composer tools menu.

Reference: OpenAI Apps SDK - connect from ChatGPT.

#Claude

Use this path for Claude chat, Claude Desktop, or Cowork custom connectors.

1. Open Claude settings.
2. Go to Customize > Connectors.
3. Click the + button next to Connectors.
4. Choose Add custom connector.
5. Enter FPF Reference as the name and the hosted endpoint as the URL.
6. Add it, then connect it in the same way as other Claude connectors.

Reference: Claude Help - custom connectors.

#VS Code

Use the MCP: Add Server command, or add a workspace config at .vscode/mcp.json:

{
  "servers": {
    "fpf_reference": {
      "type": "http",
      "url": "https://mcp.fpf.sh/api/mcp/fpf_reference/mcp"
    }
  }
}

Restart or start the server from the MCP code lens, then confirm trust when VS Code asks.

Reference: VS Code - add and manage MCP servers.

#Zed

Open the Agent Panel settings with agent: open settings, then add a custom server:

{
  "context_servers": {
    "fpf_reference": {
      "url": "https://mcp.fpf.sh/api/mcp/fpf_reference/mcp"
    }
  }
}

The indicator next to the server should turn green when the server is active.

Reference: Zed - Model Context Protocol.

#Codex CLI

Add the remote streamable HTTP server:

codex mcp add fpf_reference --url https://mcp.fpf.sh/api/mcp/fpf_reference/mcp

Equivalent ~/.codex/config.toml entry:

[mcp_servers.fpf_reference]
url = "https://mcp.fpf.sh/api/mcp/fpf_reference/mcp"

Reference: Codex configuration reference.

#Claude Code

Add the remote HTTP server:

claude mcp add --transport http fpf_reference https://mcp.fpf.sh/api/mcp/fpf_reference/mcp

Check it inside Claude Code with /mcp.

Reference: Claude Code - connect to tools via MCP.

#Pi

Pi supports MCP through extensions. Install an MCP extension, then add a streamable HTTP server config.

pi install npm:pi-mcp-extension

Global config at ~/.pi/agent/mcp.json or project config at .pi/mcp.json:

{
  "mcpServers": {
    "fpf_reference": {
      "transport": "streamable-http",
      "url": "https://mcp.fpf.sh/api/mcp/fpf_reference/mcp",
      "lifecycle": "eager"
    }
  }
}

Start Pi and use /mcp to check status.

Reference: Pi MCP extension.

#Good first prompt

Use only fpf_reference. First call get_fpf_index_status. If the index is available, find the smallest FPF route for this work: <describe work>. Return Context | Route ID | Ordered IDs | Friction avoided | Acceptance check | Next move.

Keep route answers compact. Read exact generated docs only when wording matters, and do not paste the full FPF into the chat.