Gmail Postmaster Tools MCP

Gmail Postmaster Tools MCP

Query your domains, sender compliance status, and Gmail traffic metrics from Claude, Cursor, or any MCP client using the Gmail Postmaster Tools API.

Category
Visit Server

README

Gmail Postmaster Tools MCP

License: MIT Built with MCPB Node Dependencies

An MCP server for the Gmail Postmaster Tools API (v2) — query your domains, sender compliance status, and Gmail traffic metrics from Claude, Cursor, or any MCP client.

This is a bring-your-own-credentials tool for hands-on senders: you supply your own Google OAuth client, and the server runs the sign-in locally. Nothing is hosted and no data leaves your machine.

Want this without setting up a Google Cloud project? That's what Postmaster+ is for — hosted, multi-provider (Gmail + Outlook + more), with verification, history, and alerting.

Built & maintained by the Postmaster+ team at OptiPub.

What you get

  • list_domains / get_domain — your registered domains and their verification state
  • get_compliance_status — SPF, DKIM, DMARC, alignment, message formatting, DNS records, TLS encryption, user-reported spam rate, and one-click / honored unsubscribe verdicts
  • query_domain_stats — spam rate, auth success (SPF/DKIM/DMARC), TLS encryption rate, delivery errors, and feedback-loop metrics over any date range
  • gpt_authenticate / gpt_auth_status / gpt_sign_out

Note: Postmaster Tools v2 covers traffic metrics + compliance. Domain and IP reputation are not part of v2.

Prerequisites: create a Google OAuth client (one-time, ~5 min)

You need a Google Cloud OAuth client because Google doesn't offer a shared public one for this API.

  1. Go to the Google Cloud Console and create (or pick) a project.
  2. APIs & Services → Library → search "Gmail Postmaster Tools API" → Enable.
  3. APIs & Services → OAuth consent screen (redirects to Google Auth Platform). If the project has never been configured, click Get started and set an App name + support email, Audience = External, and a contact email. If it's already configured, you'll just see the tabs — skip to the next step.
  4. Open the Audience tab. If User type = External, add your Google address under Test users → + Add users and Save (leave status as Testing — works without app verification, up to 100 users). If User type = Internal (a Workspace org project), no test users are needed — org users can sign in directly.
  5. Open the Clients tab (or APIs & Services → Credentials) → Create client → application type Desktop app → Create.
  6. Copy the Client ID and Client secret.

You'll paste those two values into this extension's configuration at install time. (Make sure the Google account you sign in with is one that has access in Gmail Postmaster Tools.)

Install in Claude / Cowork (the .mcpb)

The .mcpb is a packaging convenience for Claude Desktop / Cowork only.

  1. Download gmail-postmaster-tools.mcpb from the latest release.
  2. Settings → Capabilities → install extension, pick the file.
  3. When prompted, paste your Google OAuth Client ID and Client Secret.
  4. Run gpt_authenticate once — your browser opens for the Google sign-in.

Use with other MCP clients (Cursor, VS Code, Windsurf, …)

Underneath, this is a standard stdio MCP server — any client that runs local MCP servers can use it directly, no .mcpb required. Clone or download the repo, then point the client at server/index.js and pass your Google credentials as env vars.

Cursor — edit ~/.cursor/mcp.json (global) or .cursor/mcp.json (per project):

{
  "mcpServers": {
    "gmail-postmaster-tools": {
      "command": "node",
      "args": ["/absolute/path/to/gmail-postmaster-tools-mcp/server/index.js"],
      "env": {
        "GPT_CLIENT_ID": "xxxxx.apps.googleusercontent.com",
        "GPT_CLIENT_SECRET": "xxxxx"
      }
    }
  }
}

VS Code (.vscode/mcp.json), Windsurf, Claude Desktop (manual config), and most other clients use the same command + args + env shape — only the config file location differs. Other env overrides are listed under Configuration.

The first tool call (or gpt_authenticate) opens your browser for the Google sign-in; the 127.0.0.1 loopback works on any local desktop client.

Remote-only clients (e.g. Perplexity, ChatGPT connectors): these accept only a remote HTTPS MCP server URL, not a local command, so this stdio build can't be added directly. That would require hosting it — which is what Postmaster+ provides.

Usage examples

  • "Sign me in to Gmail Postmaster Tools" → gpt_authenticate
  • "List my Postmaster domains" → list_domains
  • "Is example.com compliant with Gmail's sender rules?" → get_compliance_status { domain: "example.com" }
  • "Spam rate for example.com over the last 30 days" → query_domain_stats { domain: "example.com", start_date: "...", end_date: "...", metrics: ["SPAM_RATE"] }
  • "DKIM auth success for example.com" → query_domain_stats { domain: "example.com", metrics: [{ "standardMetric": "AUTH_SUCCESS_RATE", "filter": "auth_type=\"dkim\"" }] }

Some metrics require a filter: AUTH_SUCCESS_RATE (auth_type=spf|dkim|dmarc), TLS_ENCRYPTION_RATE (traffic_direction=inbound|outbound), DELIVERY_ERROR_RATE (optional error_type=...). SPAM_RATE needs none.

How authentication works

OAuth 2.0 authorization-code + PKCE, over a http://127.0.0.1:<dynamic-port> loopback redirect (Google's recommended flow for desktop apps). On sign-in the server starts a short-lived listener on 127.0.0.1, opens your browser, receives the code, and exchanges it (with your client secret + PKCE verifier) for tokens.

Google requires a client secret even for desktop apps, which is why both values are needed. Tokens are stored only on your machine at ~/.gmail-postmaster-mcp/tokens.json (0600) and are git-ignored.

Scopes: by default the server requests postmaster.traffic.readonly (metrics + compliance) and postmaster.domain (list/get domains). Override with the GPT_SCOPE env var if you want to narrow it (e.g. traffic-only, dropping list_domains/get_domain).

Token lifetime (important)

Google revokes refresh tokens after 7 days for OAuth apps whose publishing status is Testing — the mode the setup steps above use to skip verification. In practice that means you'll re-run gpt_authenticate about once a week. The server handles this gracefully: when the refresh fails, tool calls tell you to re-authenticate.

To stop the weekly re-auth, set your OAuth app's publishing status to In production (Google Auth Platform → Audience → Publish app). Notes for the postmaster (sensitive) scopes:

  • Production, unverified — refresh tokens no longer expire at 7 days. You'll still click through the "Google hasn't verified this app" screen, and the app is capped at 100 users until verified. For your own single-user use, this is the simple fix.
  • Verified — submit the app for Google's OAuth verification to remove the warning and the user cap. Only worth it if you're distributing the app widely.

Configuration (env overrides)

Variable Purpose Default
GPT_CLIENT_ID Google OAuth client ID (required) —
GPT_CLIENT_SECRET Google OAuth client secret (required) —
GPT_SCOPE Space-separated OAuth scopes traffic.readonly + domain
GPT_API_BASE API base URL https://gmailpostmastertools.googleapis.com/v2
GPT_TOKEN_DIR Token cache directory ~/.gmail-postmaster-mcp
GPT_LOGIN_TIMEOUT_MS Sign-in wait 180000
GPT_REQUEST_TIMEOUT_MS Per-request timeout 60000

Build

npx @anthropic-ai/mcpb validate manifest.json
npx @anthropic-ai/mcpb pack . gmail-postmaster-tools.mcpb

Pushing a vX.Y.Z tag triggers CI to pack, attach the .mcpb to a GitHub Release, and publish to the MCP Registry.

Why zero dependencies

Only Node.js built-ins (http, https, crypto, readline) — tiny bundle, no supply-chain risk, nothing to npm install.

Contributing

See CONTRIBUTING.md.

License

MIT © OptiPub


Not affiliated with or endorsed by Google. "Gmail", "Google", and "Postmaster Tools" are trademarks of Google LLC.

Recommended Servers

playwright-mcp

playwright-mcp

A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.

Official
Featured
TypeScript
Magic Component Platform (MCP)

Magic Component Platform (MCP)

An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.

Official
Featured
Local
TypeScript
Audiense Insights MCP Server

Audiense Insights MCP Server

Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.

Official
Featured
Local
TypeScript
VeyraX MCP

VeyraX MCP

Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.

Official
Featured
Local
graphlit-mcp-server

graphlit-mcp-server

The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.

Official
Featured
TypeScript
Kagi MCP Server

Kagi MCP Server

An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.

Official
Featured
Python
E2B

E2B

Using MCP to run code via e2b.

Official
Featured
Neon Database

Neon Database

MCP server for interacting with Neon Management API and databases

Official
Featured
Exa Search

Exa Search

A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.

Official
Featured
Qdrant Server

Qdrant Server

This repository is an example of how to create a MCP server for Qdrant, a vector search engine.

Official
Featured