Vouched

Ask Claude, ChatGPT or Cursor "why did our clicks drop last month?" or "which keywords does our competitor rank for that we don't?" and it answers from real data instead of guessing.

Every tool returns the same JSON envelope. Each fact in it says where it came from, how confident it is and when it was observed, so the agent (and you) can tell a Search Console number from an index estimate.

Two ways to run it

Hosted at vouchedhq.com. Add https://vouchedhq.com/mcp as a connector in your AI client and sign in. Your own Google data is free; live market data needs a Pro or Team plan and is billed at cost. See pricing.

Self-hosted. Run the same server on your own Cloudflare account, with your own Google OAuth client and, if you want market data, your own DataForSEO key. No markup, no account with us. See docs/SELF_HOST.md.

Both run the same code and expose the same tools.

Related MCP server: sightvane-mcp

Tools

19 tools, all read-only. Nothing is ever written to Google or to any site.

Your own data (free; needs a Google connection)

Market data (DataForSEO on self-host; Pro or Team plan when hosted)

Utilities

Each tool has a page with inputs, outputs and an example response at vouchedhq.com/tools. The generated reference is docs/TOOLS.md.

What a response looks like

A trimmed get_search_performance result:

{
  "schema_version": "ofe/1.0",
  "domain": "gsc",
  "data": { "domain": "xmit.sh", "startDate": "2026-08-28", "endDate": "2026-09-24", "dimensions": ["query"] },
  "facts": [
    {
      "type": "gsc.query_performance",
      "subject": ["property:...", "keyword:any:GLOBAL:transactional email api"],
      "data": { "dimensions": { "query": "transactional email api" }, "clicks": 41, "impressions": 3120, "ctr": 0.013, "position": 8.2 },
      "provenance": {
        "source_class": "webmaster_console",
        "method": "gsc.searchAnalytics.query",
        "confidence": 1,
        "observed_at": "2026-09-25T08:00:00Z",
        "cache_hit": false
      }
    }
  ],
  "deltas": [{ "fact_type": "gsc.query_performance", "field": "clicks", "previous": 29, "current": 41 }],
  "coverage": { "returned": 25, "total": null, "scope_note": "capped at 25 rows; more are available via export_dataset" },
  "resources": [{ "uri": "mcpseo://gsc/...", "description": "Up to 1000 rows for this same query" }],
  "next_actions": []
}

deltas compare with the previous period, coverage says what the result leaves out, resources holds the rest, and next_actions suggests follow-up calls when there are useful ones. The full shape is in docs/OFE_ENVELOPE.md.

Connect a client

Hosted:

# Claude Code: the plugin (MCP server plus an SEO analysis skill)
claude plugin marketplace add muditjuneja/vouched
claude plugin install vouched@vouched

# Claude Code: the MCP server only
claude mcp add --transport http vouched https://vouchedhq.com/mcp

# Gemini CLI
gemini extensions install https://github.com/muditjuneja/vouched

In Claude or ChatGPT, add https://vouchedhq.com/mcp as a custom connector. In Cursor, add it to ~/.cursor/mcp.json:

{ "mcpServers": { "vouched": { "url": "https://vouchedhq.com/mcp" } } }

Your client opens a sign-in page the first time (standard MCP OAuth). Clients that can only send a header can use an API key from the dashboard instead.

Self-hosted, point the client at your Worker's /mcp URL with your bearer token; see docs/SELF_HOST.md.

Development

npm install
cp .dev.vars.example .dev.vars   # MCP_BEARER_TOKEN at minimum
npm run db:migrate:local
npm run dev                      # http://localhost:8787/mcp

npm test
npm run typecheck
npm run lint

Set DATAFORSEO_BASE_URL=https://sandbox.dataforseo.com while developing: DataForSEO's sandbox is free and returns dummy data in the real response shape.

Built on Cloudflare Workers with Hono, D1, R2 and KV. No Google SDK: the Google APIs are called directly.

Docs

• Self-hosting

• Architecture

• Response envelope

• Tool reference

• Hosted service internals and deploying it

Privacy

Google data is read only when you ask for it and is covered by Google's Limited Use requirements. It is never sold, used for ads or used to train AI models. See the privacy policy.

Contributing and support

Issues and pull requests are welcome. For the hosted service, email hello@vouchedhq.com.

License

MIT