# AI Agent Tool Explorer

> Index of tools AI agents can call: this site, the MCP Registry or your definitions, by capability. A searchable index of tools an AI agent can call: this platform's own tools (with API, definition and MCP names), servers in the official MCP Registry, or tool definitions you paste (MCP tools/list, OpenAI function tools, Anthropic tools). Every entry gets inferred capability tags (read, write, delete, execute, network, files, payments, messaging, secrets), a risk level, a parameter summary and a machine-readable JSON-schema definition.

Part of usefulweb.site (https://usefulweb.site/). Results are deterministic calculations from the inputs and the listed assumptions; the API returns exactly what the web page shows.

## Use it
- Human interface: https://aiagenttoolexplorer.usefulweb.site/ (interactive tool at https://aiagenttoolexplorer.usefulweb.site/tool)
- JSON API: POST https://aiagenttoolexplorer.usefulweb.site/api/v1/calculate with a JSON object of inputs (schema: https://aiagenttoolexplorer.usefulweb.site/api/v1/tool)
- OpenAPI: https://aiagenttoolexplorer.usefulweb.site/openapi.json
- MCP (Streamable HTTP): https://aiagenttoolexplorer.usefulweb.site/mcp - tool `ai_agent_tool_explorer`

## Inputs
- mode (select): What do you want to explore?
- query (text): Search - Matches names, descriptions and keywords. In paste mode it filters the pasted tools.
- includeRegistry (boolean): Also search the official MCP Registry
- toolsJson (textarea): Tool definitions JSON - MCP tools/list result, OpenAI "tools" array, Anthropic "tools" array, or one definition.
- maxResults (integer): Maximum results

## Method
- Platform tools come from this site's live registry (id, name, tagline, keywords); each links to its definition (GET /api/v1/tools/{id}: input + output JSON Schema), its calculate endpoint and its MCP tool name.
- MCP Registry: GET https://registry.modelcontextprotocol.io/v0/servers?search=… (official, self-published metadata) with links to the server's latest version record, repository and remote endpoints.
- Pasted definitions are normalised from MCP (inputSchema), OpenAI (function.parameters) and Anthropic (input_schema) shapes into the MCP tools/list shape.
- Capabilities and risk use the same rules as the MCP Server Explorer: write/delete/execute from the tool name, network/files/payments/messaging/secrets from name + description. They are inferences, not guarantees.
- Search scores each query word against name (weight 3), keywords (2) and description (1); prefixes count 60 %.