What MCP tools can do

An InstantMCP.io user MCP turns your indexed sources into Model Context Protocol tools over Streamable HTTP. Clients (Cursor, Claude, ChatGPT) call those tools to answer from your content.

Built-in tools

Every live user MCP exposes at least:

Per-source tools

When the MCP has more than one ready source, InstantMCP.io also registers one tool per source, named like ask_<source_label>. Use these when the agent should stay inside a single corpus (for example docs vs. pricing PDF).

AI-proposed tools

On MCP creation, InstantMCP.io calls an LLM to suggest additional coherent tool names and descriptions for your knowledge base (still RAG-backed). Those tools are stored on the server and exposed alongside ask.

What a tool is

Each tool has a name, a short description, and a JSON Schema for arguments (typically query). When the user asks something relevant, the client invokes the tool and uses the retrieved snippets in the reply.

Example: product docs MCP

Connect a documentation site. You typically get:

• ask: “How do I reset a webhook?”
• ask_docs / ask_changelog (if multiple sources)
• Extra AI-proposed tools such as ask_onboarding or lookup_limits

Designing good sources

• Prefer structured headings and short pages over one giant dump
• Re-sync after major content changes so embeddings stay fresh
• Use clear source labels: they become per-source tool names

Protocol notes

User MCP endpoints speak Streamable HTTP (MCP specification): POST to https://{slug}.instantmcp.link/mcp with Authorization: Bearer <api_key>, Content-Type: application/json, and Accept: application/json, text/event-stream.

Next steps

• Getting started: create your first server
• Connect: plug the MCP URL into Cursor or Claude
• FAQ: sources, hosting, and security