MCP server
Let your AI assistant search this directory directly. It gets real, sourced AI agent deployments with an evidence label and caveats, instead of guessing. Free, read-only, no sign-up, no key.
Server URL
https://agentsinthewild.co/mcpConnect
- Claude (web and desktop): Settings → Connectors → Add custom connector. Paste the server URL.
- Claude Code: run the command below.
- Other MCP clients (ChatGPT, Cursor, VS Code, and more): add a remote MCP server with the type “HTTP” (Streamable HTTP) and the server URL.
claude mcp add --transport http agents-in-the-wild https://agentsinthewild.co/mcpTools
search_use_cases: Search by words, category, industry, tool, or evidence label.get_use_case: Get one entry with its metrics, caveats, and source links.get_dataset_stats: Get the numbers behind the live report.
Try asking
- “Which banks run AI agents in production, and who reported the results?”
- “Find real coding agent deployments with independent or research evidence.”
- “What share of published AI agent results are self-reported?”
Rules
The server only reads the public directory. It stores nothing you send. Results are claims, labelled by who made them. See how we verify. The same data is also available as JSON and llms.txt.