# We shipped an MCP server: the lab's tools, inside your AI agent

> GEO/AEO Playbooks now runs an MCP server: live 5-signal site audits, llms.txt linting, knowledge-base search and the Pulse benchmark — callable natively from Claude Code, Cursor or any MCP client.

Source: https://geoplaybooks.com/blog/2026-07-12-mcp-server/

# We shipped an MCP server: the lab's tools, inside your AI agent

The lab now runs a [Model Context Protocol server](/mcp) at `https://geoplaybooks.com/mcp`. Point Claude Code, Cursor or any MCP client at it and your agent can audit a site's AI visibility, lint an llms.txt, search our playbooks and read the live Pulse benchmark — natively, as tools, without leaving the editor.

One line to install:

## Why a GEO lab ships agent tooling

Every report we sell ends with an **AI fix-it prompt** — an engineering brief you paste into an agent: findings, implementation rules, acceptance criteria. Customers loved the idea and immediately hit its limit: the agent implementing the fixes had no way to *verify* its own work. It could edit robots.txt; it couldn't re-run the audit.

Now it can. The loop closes: audit → brief → agent implements → agent calls `check_site` again → repeat until 5/5 PASS. The acceptance criteria in our reports stopped being instructions for a human and became a test suite for a machine.

## What's in the box

> The five tools check_site — live 5-signal GEO audit of any domain, 0–100 score with findings. get_fix_prompt — audit plus a ready-to-execute engineering brief. validate_llms_txt — spec lint by domain or raw content, per llmstxt.org. search_playbooks — full-text search across 30+ evergreen GEO/AEO playbooks. pulse_leaderboard — the hourly benchmark of well-known sites.

Design notes, for the technically curious: it's a stateless Streamable-HTTP endpoint — every call is one JSON-RPC request, no sessions to leak, riding the same engines that power the [free checker](/check/) and the [validator](/tools/llms-txt-validator/). Rate-limited, because `check_site` performs real fetches against real sites.

## What it costs

Nothing. The five tools are free, like the rest of the [tools page](/tools/). The paid layer comes later: Monitor subscribers will get keyed access to Share-of-Voice sampling and deep audits from inside their agents — the same deliverables the [Monitor](/monitor/) mails monthly, on demand.

If you wire it into an agent and something misbehaves, reply to any of our emails or write hello@geoplaybooks.com — bug reports from real agent runs are exactly the telemetry a practice lab wants.