MCP in Cursor settings, TypeSafe skills in the agent, no imaginary Grok-only APIs.
What Cursor provides
md) loaded for compatible agent modes. There is no separate documented Grok Bot API for Jev: when you use Grok as a model inside Cursor or related X integrations, wire Jev the same way you would for any other model (MCP tools and skills).
Register Jev MCP in Cursor
Community MCP servers document a Cursor mcpServers block. blakestone-x/jev-mcp is a common starting point (classify, score, check via uvx). Set TYPESAFE_API_KEY in the env block or your shell.
{
"mcpServers": {
"jev": {
"command": "uvx",
"args": ["--from", "git+https://github.com/blakestone-x/jev-mcp@v0.2.1", "jev-mcp"],
"env": { "TYPESAFE_API_KEY": "..." }
}
}
}Agent skills in Cursor
Install TypeSafe's official skill so the agent learns System One patterns before calling tools:
npx skills add typesafe-ai/skills --skill typesafe-ai
# Optional judge playbook:
npx skills add gecm0/jev-judge-mcp --skill jevjevwire and other MCP toolkits
Brainwires/jevwire documents multiple tools (jev_evaluate, jev_rank, jev_verify, jev_gate_action) and npm package jevwire for Streamable MCP. rashedInt32/jev-mcp and burnigtm/jev-mcp are alternate servers listed in the directory; compare tool surfaces before picking one.
Grok in the loop
Demos such as pixelml-av-grok-jev show Grok generating answers while Jev filters evidence and checks support. That pattern is application-level (RAG plus refiner), not a Cursor-built Grok feature.
In Cursor, you might use Grok as the chat model while Jev MCP tools handle decide steps; keep responsibilities split.
Cloud Agents and keys
Cloud Agents need secrets in the environment the remote worker uses, not only your laptop shell. Follow Cursor documentation for MCP and env injection on cloud runs. Rotate keys if they ever appear in committed mcp.json.
Environment keys
ai/ (or route through Vercel AI Gateway / OpenRouter where the integration documents it). Export TYPESAFE_API_KEY in the environment for Cursor MCP and agent processes.
Some MCP clients accept AI_GATEWAY_API_KEY when using gateway-hosted Jev models; follow that integration's README. Never commit keys into repos or paste them into chat logs.
export TYPESAFE_API_KEY="tsk_..."
# Optional pin:
export TYPESAFE_MODEL="jev-latest"Pitfalls
Pin MCP package versions (git tags or npm versions) so builds do not drift. MCP tools are not automatic: the model must call them unless you use a separate hook-based plugin on Claude Code.
Avoid stuffing every Jev MCP schema into context when you only need judge; pick one server. Read each server's license and data handling before sending production content.
Related directory listings
Related agent guides
FAQ
- Is there a Grok Bot-only Jev SDK?
- We did not find one. Use standard TypeSafe credentials and documented MCP or HTTP APIs regardless of which chat model Cursor selects.
- Cursor vs Claude Code for jev-judge?
- jev-judge-mcp targets Claude Code, Codex, and OpenCode explicitly. Cursor users typically register the same stdio server manually in MCP settings.
- Which jev-mcp fork?
- blakestone-x/jev-mcp documents Cursor config. arunav25 and rashedInt32 forks add benchmarking or extra tools; read README diffs on the catalog detail pages.
- Skills vs MCP in Cursor?
- Skills steer prompting and workflow design. MCP executes typed calls. Use both when the agent needs to know when to judge and a tool to do it.
- Gateway instead of direct TypeSafe?
- See /learn/vercel-ai-gateway for AI_GATEWAY_API_KEY and model ids when your stack already routes through Vercel.