Every MCP-capable host — Claude Code, Claude Desktop, Cursor, Codex, and the rest — can read your sites' scores, audits, Search Console data, keywords, competitors, authority, and AI visibility as 40 first-class tools, backed by the same versioned contract as the REST API.
Streamable HTTP JSON-RPC. Two ways to authenticate:
OAuth 2.0 + PKCE — clients discover it automatically via /.well-known/oauth-protected-resource and walk you through consent in the browser.
API key — Authorization: Bearer tg_live_..., created in Account → Keys. read scope covers all reads; write adds audit triggering.
Connect from your client
Two ways in. Smithery gives you guided install flows for a dozen hosts from one listing. Direct setup is a single command or config block against https://trustgrowth.ai/mcp.
Option 1 · Via Smithery
One listing with click-through install for Claude, Cursor, VS Code, Windsurf, ChatGPT, and more — plus per-profile key storage, so hosts without their own connector UI still get in.
Point your client at the endpoint. OAuth consent opens in the browser on first use, or paste an API key from Account → Keys. Exact steps per client below.
Docs are public; API and MCP requests require a Starter plan or higher. Create and manage keys in Account → Keys.
What the tools cover
Scores & history
Daily site scores, pillar breakdowns, time-series snapshots, and the publication-safe evidence packet for numbers that leave the building.
Audits with a verify loop
Open issues by severity, first-seen and fixed history, and a write-scoped trigger to queue a fresh audit and poll it to completion.
Search Console reads
Per-query and per-page daily rows, index verdicts, and SERP snapshots — persisted, credit-free reads that never trigger a fetch.
Keywords & competitors
Keyword lifecycle (quick wins, striking distance, content gaps), per-keyword history, and competitor observation history.
Authority & links
The authority pillar, paginated referring domains, link review marks, and the prospect pipeline with write-scoped updates.
AI visibility
The five-stage funnel — reach, readable, retrieved, recalled, impact — with each stage's evidence labeled deterministic, census, sample, or observed.
Every tool responds with the same versioned envelope as REST; the contract version rides in meta.contract_version, and read tools are annotated read-only so hosts can run them without confirmation prompts. Tool schemas, annotations, and the full catalog live in the server card.
Use TrustGrowth from your AI chats
Once connected, your growth data answers in plain language — no dashboard tab required.
In a chat · MCP
Ask your live numbers a question
With the connector added, ask Claude — or any chat that accepts MCP connectors:
"How did my site's score move this week, and which issues opened?"
The model calls get_site_score and list_site_issues and answers from your measured history — every read tool is annotated read-only, so nothing runs behind your back.
In a coding agent · Skills
Groundcrew: 20 open-source skills
Prefer plain agent skills over a tool server? Groundcrew wraps this surface in 20 MIT-licensed Markdown skills — audits, keyword research, standups, and a survey → fix → verify → report loop:
Then ask your agent to "run a growth standup for my site" — skills work with public evidence alone, and connect to the same API when you add a key.
Frequently asked questions
What is the TrustGrowth MCP server?
A hosted Model Context Protocol server at https://trustgrowth.ai/mcp. It exposes your sites' scores, audits, Search Console reads, keywords, competitors, authority, and AI visibility as first-class tools over streamable HTTP, backed by the same versioned contract as the REST API.
Which AI clients can connect?
Any MCP-capable host: claude.ai and Claude Desktop through a custom connector, Claude Code, Cursor, VS Code, and Codex, plus the hosts Smithery covers with its guided install flows. The connect section above has exact steps per client.
How does authentication work?
Two ways. OAuth 2.0 with PKCE: clients discover it automatically and open TrustGrowth's consent screen in the browser. Or an API key created in Account → Keys, sent as a bearer token; the read scope covers every read tool, and write-scoped tools need the write scope.
Can a connected AI change my site?
No. The tools read measured evidence; nothing edits your site, your content, or your repository. Read tools are annotated read-only so hosts can run them without confirmation prompts, and the few write-scoped tools do bounded product actions like queueing a fresh audit.
Do read tools trigger crawls or spend credits?
No. Reads return persisted evidence — audits, Search Console rows, snapshots already collected on your plan's schedule — and never start a fetch. Triggering a fresh audit is a separate write-scoped tool you call deliberately.
What does it cost?
The documentation is public; API and MCP requests require a Starter plan or higher and share your plan's API limits. There is no separate MCP fee.
How is this different from Groundcrew skills?
MCP is a live tool server: your AI client calls TrustGrowth directly and answers from measured data. Groundcrew is an open-source pack of 20 Markdown skills that needs no account and works from public evidence — and uses the same API when you connect a key. Use MCP in chat apps, Groundcrew in coding agents, or both.
How do I revoke access?
In Account → Keys: delete an API key or revoke an OAuth grant, and every client using it loses access immediately. Grants and keys are scoped to your account and only reach the sites it owns.