One command gives Claude — or any MCP-capable agent — read-only access to your Code Results account: Search Console performance annotated with your merged PRs, plus anonymized benchmarks. No CSV exports, no copy-paste.
Ask a chat window why your rankings dropped and you get generic advice — the same checklist it would give anyone. The model has no idea what your team shipped or when Google reacted. Feeding it data means exporting CSVs from Search Console, pasting them into the conversation, and hoping the context survives.
The Code Results MCP server replaces that ritual. Your agent queries your account directly: search performance lined up against your merged PR timeline, calibrated against anonymized cross-customer benchmarks. The advice stops being generic because the evidence stops being generic — it's your deploys, your queries, your lag between merge and movement.
$ claude mcp add --transport http code-results https://code-results.com/mcpOn first use, the agent opens your browser. You sign in at code-results.com, and a consent screen shows exactly what the agent gets — read-only access to your account's search data, PR timeline, and benchmarks. Approve it and you're done. That's the whole setup.
The command above is for Claude Code, but the server speaks standard streamable HTTP — any MCP-capable agent can connect the same way.
The server exposes five tools. Nothing hidden, nothing writable.
See which sites are connected, which repositories are paired with them, and when each last synced.
Daily clicks, impressions, and impression-weighted average position for a site — up to 120 days back.
The queries driving your traffic, ranked by clicks, with CTR and position for each — up to 200 at a time.
Merged PRs across your tracked repositories with dates and change sizes — up to a year back. This is the half of the picture Search Console can't give you.
Anonymized cross-customer baselines: how often shipped PRs improve attributed positions, how large the strong improvements run, and the typical lag between merge and ranking movement.
Together they let an agent do real diagnosis. Because Google's crawl-and-index lag runs days to weeks, the PR that caused a shift merged before the metric moved — an agent with both the timeline and the daily data can walk that window instead of guessing.
Tools give an agent access; skills give it a method. We publish three, each a single markdown file you save from code-results.com/skills/<name>.md into your agent's skills directory.
Turns the week's search data and merged PRs into a stakeholder-ready summary of what moved and what shipped.
When a metric drops, walks the PR timeline back through the crawl-lag window to surface suspect deploys — and distinguishes ranking changes from demand changes from snippet changes.
Ties engineering work to search outcomes in language a budget owner accepts — evidence-first, honest about uncertainty.
It's read-only. The agent can query your data. It cannot change settings, modify code, or touch billing — those capabilities don't exist on the server.
It's scoped to your account. Your agent sees your properties and your repositories, nothing else. Benchmarks are anonymized aggregates from the cross-customer pool — no individual customer's data is exposed, and the pool's size and composition are never disclosed.
You can revoke it anytime. Settings → Connected AI agents. Disconnecting revokes the agent's sessions and refresh tokens.
Attribution is evidence, not proof. The attribution model is probabilistic — a good agent treats it that way, and the skills are written to say so.
Create a free account, connect GitHub and Search Console, then run the install command. Your agent goes from generic SEO advice to reasoning over your actual deploys.
Start for free