Connecting Claude Code to an SEO MCP server, and what a safe first session looks like

Description and use

Connecting Claude Code to the Crawl Foundry SEO MCP server enables real keyword data retrieval and budget control, preventing accidental purchases.

Problem

Claude models generate fabricated keyword search volumes, risking reliance on inaccurate data.

Outcome

Agents now retrieve accurate SEO data, avoid accidental purchases, and operate within defined budgets.

Workflow

• Add the MCP server with claude mcp add --transport http --scope user crawl_foundry "https://app.crawlfoundry.com/api/mcp"
• Authenticate via OAuth when the agent first calls the server
• Confirm the workspace by asking the agent to "Show me the saved SEO overview for this workspace. Don't buy new data and don't change anything."
• Review the returned data gaps and coverage states
• Request a paid action with a cost preview, e.g., "Research keywords around "self-hosted analytics". Show me the expected cost before you run anything."
• Verify the agent’s budget limits (EUR 5 per action, EUR 10 per workspace/month, EUR 50 per organization/month)
• After the session, run claude mcp list to confirm connection
• Check the activity and cost log in Crawl Foundry for paid runs
• Revoke and reconnect to test quick access revocation

Evidence

Ask a coding agent which keywords your docs should target and it will happily give you a list with search volumes. The volumes are invented.

Source

The Model Context Protocol (MCP) is the standard way to fix that: the agent calls tools on a server that has the real data.

Source

The first time the agent calls the server, your client opens the Crawl Foundry sign-in in the browser. You log in, see which workspace and permissions are requested and allow access.

Source

Show me the saved SEO overview for this workspace. Don't buy new data and don't change anything.

Source