Expose evaluate_agent_quality, test_rfp_response, test_io_execution as direct MCP tools on /mcp endpoint

[tescoboy]

Problem

The /mcp endpoint currently exposes discovery and registry tools directly, but higher-value evaluation tools are only available via the chat_with_addie conversational interface — not as callable MCP tools.

This means external clients (Claude Code, Claude Desktop, custom agents) cannot call these tools directly. When Addie tries to call them internally in chat_with_addie, it fails silently and falls back to "I can't
run that."

Tools to expose directly

• evaluate_agent_quality — protocol compliance evaluation with per-track results
• test_rfp_response — test how a publisher's agent responds to an RFP brief
• test_io_execution — map IO line items to agent products and generate create_media_buy JSON
• probe_adcp_agent — check if an agent is online and responding

Impact

Developers building buyer agents or testing seller agents via Claude Code/Desktop currently get an incomplete picture of what's available. Addie's chat over-describes what the MCP endpoint can do, which erodes trust.

Proposed solution

Implement these as first-class MCP tools on the /mcp endpoint with the same parameters documented in the AdCP spec and Addie's internal tool definitions.

Affected files

• MCP endpoint tool registration (server-side)
• /mcp tool manifest