MCP Proxy

The MCP Proxy is a transparent HTTP forwarder that bridges your application to the underlying Jiro search engine's MCP-over-HTTP server. It accepts any HTTP method (GET, POST, PUT, DELETE, etc.) and forwards the request body, method, and path to the upstream MCP server.

The MCP (Model Context Protocol) is a standard protocol for AI agents and tools to interact with external services. Jiro's MCP server exposes search, scrape, and AI capabilities as MCP tools that can be used by AI coding assistants like Claude Desktop, Cursor, OpenCode, and others.

This proxy endpoint is used when: - Your AI agent needs to communicate with Jiro via MCP over HTTP - You're building custom MCP integrations - You need to forward MCP requests through the Jiro API gateway

The response is streamed back chunk-by-chunk for real-time output from AI agents.

Each MCP proxy request costs 3 credits. Credits are deducted before the upstream call is made. If the upstream call fails (timeout, connection error, non-2xx response), credits are automatically refunded.

The 3-credit cost applies regardless of: - The HTTP method used (GET, POST, etc.) - The path being called - The size of the request or response - Whether the upstream succeeds or fails

The MCP proxy follows a 5-step process:

1. 1Request received — your request hits /v1/mcp/{path} with any HTTP method
2. 2Credit deduction — 3 credits are deducted from your balance. If you have insufficient credits, the request is rejected with a 402 error before reaching the upstream
3. 3Forward to upstream — the raw body, headers, and method are forwarded to the Jiro MCP server at 127.0.0.1:8080/mcp/{path}
4. 4Stream response — the upstream response is streamed back to you as-is, preserving the original format and content
5. 5Refund on failure — if the upstream call fails (timeout, DNS error, connection refused, non-2xx status), the 3 credits are automatically refunded to your balance

Forward any HTTP method to any MCP path:

The Jiro MCP server exposes the following tool paths:

• /v1/mcp/search — search the web via MCP protocol
• /v1/mcp/scrape — extract content from URLs
• /v1/mcp/ai_search — AI-powered multi-step search
• /v1/mcp/ai_synthesize — synthesise content from context
• /v1/mcp/engines — check engine status
• /v1/mcp/status — check server status and credits
• /v1/mcp/tools — list all available MCP tools and their schemas

Each path accepts the same parameters as the corresponding REST API endpoint. The MCP protocol adds a JSON-RPC layer on top of the HTTP request.