MCP Server Setup

The Revnu MCP server connects Claude, ChatGPT, Cursor and other MCP hosts to your Revnu growth agent. Its tools map one-to-one onto the API's routes, so any MCP host can read your queue, approve or send back a card, talk to your agent, label leads, manage the block list and edit workflows. Connect it two ways:

• Remote, at https://revnu.com/api/mcp. Nothing to install. You sign in to Revnu once and pick what the app may do.

• Local, as revnu mcp from the @revnu/cli package, with an API token in its environment.

Both serve the same tools. On hosts that support MCP Apps, review_list and review_get open your review queue as cards with Approve, Send, Request Changes, Snooze and Dismiss buttons, right in the chat.

Remote connector for Claude, ChatGPT and Cursor

The server URL is:

https://revnu.com/api/mcp

It speaks Streamable HTTP and signs you in with OAuth. The first time you connect, Revnu asks you to sign in and choose which permissions the app gets. With no Revnu account yet, connecting creates one, and the app sets it up with you through the setup_* tools (revnu.com/setup). That creates an API token labelled with the app's name, which you can revoke any time in Settings → Developers. An app that goes 90 days without using its token is disconnected; connect it again to get a new one.

Claude and Claude Desktop

Settings → Connectors → Add custom connector, paste the URL, then Connect.

ChatGPT

In ChatGPT's connector settings, turn on developer mode, create a connector with the URL above and choose OAuth as the authentication.

Claude Code

claude mcp add --transport http revnu https://revnu.com/api/mcp

Then run /mcp inside Claude Code to sign in.

VS Code

In .vscode/mcp.json:

{
  "servers": {
    "revnu": { "type": "http", "url": "https://revnu.com/api/mcp" }
  }
}

Cursor and other hosts

{
  "mcpServers": {
    "revnu": { "url": "https://revnu.com/api/mcp" }
  }
}

A host that can't run OAuth can send a token instead: add the header Authorization: Bearer rvp_... with a token from Settings → Developers.

Local server

Claude Desktop and Cursor

Add to claude_desktop_config.json or Cursor's MCP settings:

{
  "mcpServers": {
    "revnu": {
      "command": "npx",
      "args": ["-y", "@revnu/cli", "mcp"],
      "env": { "REVNU_API_TOKEN": "rvp_..." }
    }
  }
}

Claude Code

claude mcp add revnu -e REVNU_API_TOKEN=rvp_... -- npx -y @revnu/cli mcp

Tools

Start with ask_revnu. It asks your growth agent to do work (find and research leads, draft outreach, write content, change ads, report results) and waits for the answer, which can take a minute or two. Hosts that support progress updates show that the agent is still working. If the wait ends first, it returns queued or running with a requestId: check it with agent_message_status. While that request is unfinished, ask_revnu does not send another message, because a new message would stop the running work. It waits on the earlier request instead and says so, unless the call passes interrupt: true to replace that work or queue: true to have the agent start the new job once the current one is done. Anything the agent wants approved lands in your review queue.

The rest read or change your account directly: whoami, review_list, review_get, review_answer, review_send, review_request_changes, review_snooze, review_dismiss, agent_message_status, agent_thread, agent_status, results_get, workflows_list, workflow_runs, workflow_create, workflow_update, workflow_pause, workflow_resume, workflow_run, workflow_delete, leads_list, lead_get, lead_update, linkedin_person, blocklist_list, blocklist_add, blocklist_remove.

While an account is still in setup it also sees setup_status, setup_website, setup_confirm, setup_checkout and setup_finish, and the other tools answer setup_pending until the trial starts. Your assistant only sees the tools its token's scopes allow. Connect a set-up account with only Review: Read ticked and it sees whoami, review_list and review_get, nothing else.

Careful. Writes through an assistant are real writes: review_answer on a blog draft publishes it, review_send sends email, blocklist_remove lets the agent contact that address again, and workflow_delete cannot be undone. Mint a read-only token if your assistant should only look.