Jev in Cursor

Cursor speaks MCP, so Jev drops in as a decision tool. Add the open jev-mcp server and Cursor's agent can call classify / score / check / gate / decide mid-task — typed, calibrated answers in milliseconds, not a slow, unparseable second model call.

Cursor's agent makes a lot of small structured decisions while it works — which file, is this edit safe, does this test failure matter, which approach wins. Done in free-text reasoning those are slow and inconsistent. Jev gives Cursor a typed decision tool it can call and branch on directly, so the agent reserves its model budget for actual coding.

Add Jev to Cursor

Cursor reads MCP servers from .cursor/mcp.json in your project (or ~/.cursor/mcp.json globally). Add jev-mcp — it runs straight from GitHub with npx, no clone or build:

{
  "mcpServers": {
    "jev": {
      "command": "npx",
      "args": ["-y", "github:codaaiteam/jev-mcp"],
      "env": { "TYPESAFE_API_KEY": "jv_live_..." }
    }
  }
}

Reload Cursor and enable the server in Settings → MCP. The Jev tools appear in the agent's toolset. A hosted jv_live_ key routes to the gateway automatically — no endpoint to configure.

What the agent can now call

• jev_classify — put an input into one of your categories
• jev_score — rate on an ordered scale you define
• jev_check — a calibrated yes/no (noul) with a probability
• jev_gate — allow / confirm / block a risky action before it runs
• jev_decide — many typed questions in one round trip

Because each answer is typed and calibrated, Cursor branches on a probability instead of re-reading its own prose — faster loops, fewer format mistakes, and a real threshold you can set for when to auto-proceed vs. ask you.

Config reference

The key and endpoint details — including the hosted-key gotcha — are on the Jev MCP page, since Cursor uses the same jev-mcp server as every other MCP client.

FAQ

See also: Jev MCP server (config) · Claude Code integration · Agent skill (all clients)