Model Context Protocol is an open way for AI assistants like Claude to subscribe to external data sources and tools. WebmasterID Agent exposes an MCP surface so Claude can read your recommendations, site context, and repository mappings without scraping.

Does Claude deploy code through MCP?

No. MCP gives Claude read access to Agent data and the ability to prepare prompts. Code changes are reviewed and deployed by the operator — there is no autonomous deploy and no auto-merge.

What data can Claude inspect through MCP?

Recommendations, repository mappings, site metadata, AI-crawler activity, AI referral attribution, and execution lifecycle history. Everything is workspace-scoped — Claude only sees the workspace you connect.

What tasks can Claude prepare?

Operator-approved improvements derived from real Agent recommendations: heading-structure cleanups, internal-link additions, llms.txt updates, install-snippet fixes, metadata adjustments. Claude prepares the diff; you review and deploy.

Will MCP improve my search rankings?

We don't promise ranking outcomes — no tool can. MCP makes it faster to take grounded operational action on what your data actually shows. The work is honest; the outcomes depend on many factors outside any single tool's control.

MCP — live on Pro

Connect Claude to your WebmasterID Agent

WebmasterID Agent ships a Model Context Protocol server that Claude reads from over HTTP. Workspace-scoped Bearer auth via the api_keys table; ten read-only tools (sites, AI referrals, bot intelligence, recommendations, repo mappings, execution sessions, plan status). Operator approval required for every change — no autonomous deploy.

Start with Pro See pricing

What is MCP

A standard surface Claude can read from

Model Context Protocol is an open protocol for letting AI assistants subscribe to external data sources and tools. WebmasterID ships a real MCP HTTP server at /api/agent/mcp on the dashboard host — JSON-RPC 2.0, workspace-scoped Bearer auth, ten read-only tools. No scraping, no screen-reading, no autonomous execution.

What Claude can call

Ten read-only tools, today

Every MCP tool returns compact, safe-projected data. No raw IPs, no user-agents, no secrets, no prompt bodies. The full list:

get_sites — list workspace sites
get_site_summary — 30-day site totals
get_recent_events — safe event projection
get_bot_intelligence — AI / search / automation bots
get_ai_referrals — by AI source
get_top_pages — pageview leaderboard
get_recommendations — agent action drafts
get_repository_mappings — framework + branch + paths
get_execution_sessions — execution lifecycle
get_plan_status — current plan + features

How Claude connects

Open the MCP server, point Claude at it

On Pro or above, the Agent's MCP endpoint accepts authenticated callers at /api/agent/mcp. Configure Claude (Desktop, Claude Code, or VS Code) with the endpoint + a Bearer token from /settings/api-keys; Claude can now read your workspace's data.

Sign up on Pro or above; the Agent is included.
Generate an Agent API token from /settings/api-keys (workspace-scoped, revocable).
Add the WebmasterID MCP server to your Claude configuration.
Claude can now read recommendations, repo mappings, AI-crawler activity, and site context for that workspace.

Data Claude can inspect

What's on the MCP surface

The MCP allow-list mirrors the dashboard's read surface. Sensitive fields (raw IPs, secret keys, plaintext invite tokens) are NEVER on this surface — they aren't on the dashboard either.

· Recommendations the operator has surfaced
· Repository mappings (org / repo / framework / branch / inferred files)
· AI-crawler activity (last 30 days, per pathname)
· AI referral attribution per source
· Site metadata (URL, canonical, sitemap path, robots path)
· Execution lifecycle history

Tasks Claude can prepare

Operator-approved improvements, not autonomous edits

Examples of what Claude prepares once it has Agent data. In every case, Claude produces a diff; the operator reviews and deploys.

Heading structure
Claude reads the recommendation, opens the file referenced in the repo mapping, and proposes a heading-tree change. You review.
Internal links
Add missing internal links based on Core's observed pageview flows. Claude prepares the change with exact target paths.
llms.txt policy
Update your llms.txt and re-audit which AI crawlers respect it. Claude prepares the file; you review the policy.
Install-snippet fixes
When the tracker isn't firing on a page, Claude can prepare the snippet-placement fix based on the framework binding.

Safety

Human approval is the only deploy path

The Agent helps prepare and control website improvements with Claude. It does not deploy changes without operator approval. There is no auto-merge, no auto-deploy, no autonomous git operation.

Use cases

Why this matters for AI-search operators

If you run sites that AI assistants now read and cite, the operational work is no longer just classic SEO. The MCP surface lets Claude do the prep work grounded in your actual data, then hand it back for your review.

Indie founders who want Claude to draft the fix from real data.
SEO ecosystems running improvements across many sites at once.
Operators who want to keep humans in the loop on every change.
AI-native sites where the data that matters lives across Core + the AI-crawler view.

FAQ

Frequently asked

What is MCP?: Model Context Protocol is an open way for AI assistants like Claude to subscribe to external data sources and tools. WebmasterID Agent exposes an MCP surface so Claude can read your recommendations, site context, and repository mappings without scraping.
Does Claude deploy code through MCP?: No. MCP gives Claude read access to Agent data and the ability to prepare prompts. Code changes are reviewed and deployed by the operator — there is no autonomous deploy and no auto-merge.
What data can Claude inspect through MCP?: Recommendations, repository mappings, site metadata, AI-crawler activity, AI referral attribution, and execution lifecycle history. Everything is workspace-scoped — Claude only sees the workspace you connect.
What tasks can Claude prepare?: Operator-approved improvements derived from real Agent recommendations: heading-structure cleanups, internal-link additions, llms.txt updates, install-snippet fixes, metadata adjustments. Claude prepares the diff; you review and deploy.
Will MCP improve my search rankings?: We don't promise ranking outcomes — no tool can. MCP makes it faster to take grounded operational action on what your data actually shows. The work is honest; the outcomes depend on many factors outside any single tool's control.

Ecosystem

Built for operators

WebmasterID Agent + MCP composes with the rest of the operator stack.

Pocket Manager also on Google Play.