MCP server

Flow Relay speaks the Model Context Protocol in two ways, exposing the same 15 tools over the same REST API:

Remote server at https://www.flowrelay.it/mcp (Streamable HTTP) – nothing to install, works from any MCP client that supports remote servers.
Local server via the @flowrelay/mcp-server npm package – a thin stdio client installed on demand via npx, for clients that only speak stdio.

Remote server

The endpoint is https://www.flowrelay.it/mcp. Authenticate with your fr_ API key as a bearer token, or through the built-in OAuth flow if your client supports it. Setup differs per client – pick yours below.

Claude Desktop and claude.ai (custom connector)

No manual header and no OAuth credentials needed. In Settings > Connectors > Add custom connector:

Name: Flow Relay.
URL: https://www.flowrelay.it/mcp.
Leave the advanced OAuth Client ID and Client Secret fields empty – Flow Relay handles client registration automatically.
Click Add, then Connect: the Flow Relay authorize page opens in your browser and you paste one of your API keys once to grant access.

The connection carries exactly the permissions of that key and stops working the moment you revoke the key in Settings > API Keys.

Claude Code

bash

claude mcp add --transport http flowrelay https://www.flowrelay.it/mcp \
  --header "Authorization: Bearer fr_your_api_key_here"

Cursor, Windsurf and other JSON-config clients

Any client that takes a JSON config with remote servers:

json

{
  "mcpServers": {
    "flowrelay": {
      "url": "https://www.flowrelay.it/mcp",
      "headers": {
        "Authorization": "Bearer fr_your_api_key_here"
      }
    }
  }
}

Stdio-only clients

Bridge with mcp-remote:

json

{
  "mcpServers": {
    "flowrelay": {
      "command": "npx",
      "args": [
        "-y", "mcp-remote", "https://www.flowrelay.it/mcp",
        "--header", "Authorization: Bearer fr_your_api_key_here"
      ]
    }
  }
}

Picking a project

The connection itself is project-agnostic: one key sees every project it has access to. The model picks a project in conversation – list_projects then set_active_project – or passes project_id explicitly to any tool. The remote server keeps the active project per API key for 30 days, so it survives across sessions and devices using the same key.

Local server (npm package)

The @flowrelay/mcp-server package runs locally over stdio. No repository access is required; the server is installed on demand via npx.

Setup

Install Node.js (v22 or later).
Create an API key in Settings > API Keys.
Add the server to your Claude config file:

json

{
  "mcpServers": {
    "flowrelay": {
      "command": "npx",
      "args": ["-y", "@flowrelay/mcp-server"],
      "env": {
        "FLOWRELAY_API_KEY": "fr_your_api_key_here",
        "FLOWRELAY_PROJECT_ID": "your_project_id"
      }
    }
  }
}

Environment variables

Variable	Required	Purpose
`FLOWRELAY_API_KEY`	Yes	Your `fr_` API key.
`FLOWRELAY_PROJECT_ID`	No	Initial active project for project-scoped tools. `set_active_project` overrides it for the session; useful when the server always works on one project.
`FLOWRELAY_BASE_URL`	No	Override the API base (defaults to `https://www.flowrelay.it`).

Tools

Both servers expose the same 15 tools:

Tool	Description
`get_workspace_context`	Inspect account mode, active scope and current role.
`list_projects`	List personal and organization projects you can access.
`set_active_project`	Set or clear the active project context.
`list_filter_options`	List selectable resources, branches, event types and priorities per source.
`list_handoffs`	List handoffs by status and limit.
`generate_handoff`	Generate a handoff with optional source filters.
`list_integrations`	List connected integrations in personal or project scope.
`list_untracked_resources`	List active resources not yet assigned to a project.
`list_events`	Browse recent events in personal or project scope.
`discord_list_channels`	List text channels in your connected Discord server.
`discord_send_message`	Send a Discord message – plain text, or a handoff/insight (by id or `last_*`) rendered to Markdown and attached as a `.md` file.
`generate_correlation_insight`	Generate a cross-source correlation insight.
`generate_onboarding_brief`	Generate an onboarding brief insight.
`generate_architecture_insight`	Generate an architecture insight.
`list_insights`	List project AI insights.

The generate_* tools and list_filter_options accept per-source filters (projects / eventTypes / branches / priorities), so the model works with real selectable values rather than guessing. See Filters.

The package is published on npm under the MIT license.