Using MCP for tool discovery without surrendering judgment

Connect to the read-only endpoint

Use the live read-only Streamable HTTP endpoint: https://toolfound.com/api/chatgpt/mcp. It needs no account or API key. The complete MCP setup page at https://toolfound.com/mcp also documents Toolfound’s separate full endpoint and its email-verified write tools, so check which endpoint your client is using before any call.

• toolfound_search_tools accepts query.
• toolfound_get_product accepts slug.
• toolfound_find_by_task accepts task and optional pricing: any, free, or free_or_freemium.
• toolfound_inspect_public_link accepts sourceUrl and targetUrl, both public HTTPS URLs.

Make one schema-valid request

An MCP client normally constructs the call after listing the available tools. To exercise the endpoint directly, send the JSON-RPC request below. The task field is required, pricing is optional, and the schema rejects extra fields.

Read evidence without inventing certainty

Task matches report the listing terms that matched and preserve unknown capabilities as unknown. Public-link inspection reports anchors in one fetched HTML response; it cannot establish indexing, crawler treatment, traffic, or ranking effects. Use canonical result and listing URLs when checking or citing an answer.

Keep submission, publication, messages, purchases, and destructive changes behind explicit review. If you later choose the full endpoint documented on /mcp, confirm the target and arguments at the moment of action.

Limitations

• A protocol standardises communication; it does not certify a server, its data, or its authorization model.
• The exact confirmation and permission experience depends on the MCP client and server you use.

Primary sources

• Model Context Protocol: architecture
• Model Context Protocol: tools