MCP overview
A one-tool Model Context Protocol server exposing solvemax to any MCP-aware AI client.
solvemax-mcp is a thin Model Context Protocol server — a
fetch wrapper over the same public API described in API reference. It carries no
prompts, weights, or pipeline logic of its own; the proprietary engine runs server-side only,
exactly as it does for direct API calls.
What it gives an AI assistant
One tool, get_verdict, that any MCP-compatible client (Claude Desktop, Claude Code, Cursor) can
call mid-conversation to run a real decision through solvemax and get a formatted verdict back
inline — headline, win rate, podium, verification outcome, exit plan, trade-off, and a link to
the full slip.
Where the logic actually runs
Every call from the MCP server is a single POST to /api/v1/verdict on your solvemax
deployment (see The verdict endpoint) — the same endpoint, same
credit cost, same pipeline as any other API caller. The MCP layer only formats the request and
the response for a chat context.
Requirements
- A solvemax API key (
smx_live_...) — see Authentication. - An MCP-compatible client.
See Setup to install and configure it, and Tools for the
full get_verdict schema.
Run a decision through the full engine and get the verdict slip POST
Synchronous — the request does not return until the pipeline (clarify → deep run → verify → assemble) has finished or the 300s server budget is exhausted. On any failure the spent credit is refunded automatically (see failRunAndRefund in the source).
Setup
Configure solvemax-mcp in Claude Desktop, Claude Code, or Cursor.