Jev MCP server: typed decisions as tools for any agent

What the server is

The endpoint is https://jevstation.com/api/mcp. It speaks the MCP Streamable HTTP transport in stateless mode:

• POST only. Each request is one JSON-RPC message, or a batch of up to 20. Replies come back as application/json. GET and DELETE return 405.
• No session, no stream. The server never opens an SSE stream and does not issue a session id.
• Tools only. It implements initialize, ping, tools/list and tools/call. There are no resources, prompts or sampling.
• API key auth. Every request must carry Authorization: Bearer <key>. OAuth is not supported.
• Protocol versions 2025-06-18, 2025-03-26 and 2024-11-05 are accepted.

Behind the tools is the same code path as POST /api/v1/systemone, so answers, credits, refunds and limits are identical. See the API docs for the full answer format.

The five tools

state can be plain text or a JSON object or array, up to 24,000 characters. questions is a map of id to question, up to 12 per call:

{
  "state": "I was charged twice and support has not replied in 3 days.",
  "questions": {
    "team": {
      "type": "choice",
      "instructions": "Which team should handle this ticket?",
      "criteria": {
        "billing": "Payments and refunds",
        "tech": "Bugs and outages"
      }
    },
    "churn_risk": {
      "type": "noul",
      "instructions": "Is the customer at risk of churning?"
    }
  }
}

jev_route_model and jev_guard_tool_call use fixed question sets — the same ones behind the LLM router and tool-call guard pages — so the agent only supplies the content.

Every successful call returns the result twice: as JSON text for the model to read, and as structuredContent for clients that parse it. Both contain questions (the set that was asked), model, answers, usage, credits (charged and remaining) and latency_ms.

Connect a client

Create a key at Settings › API Keys and export it as JEVSTATION_API_KEY. The snippets below use syntax checked against each client’s official docs on 2026-09-28.

Claude Code — one command, stored in your user or local config:

claude mcp add --transport http jevstation https://jevstation.com/api/mcp \
  --header "Authorization: Bearer $JEVSTATION_API_KEY"

Or commit a project .mcp.json that reads the key from each developer’s environment:

{
  "mcpServers": {
    "jevstation": {
      "type": "http",
      "url": "https://jevstation.com/api/mcp",
      "headers": { "Authorization": "Bearer ${JEVSTATION_API_KEY}" }
    }
  }
}

More detail in Jev in Claude Code.

Cursor — ~/.cursor/mcp.json (all projects) or .cursor/mcp.json (one project):

{
  "mcpServers": {
    "jevstation": {
      "url": "https://jevstation.com/api/mcp",
      "headers": { "Authorization": "Bearer ${env:JEVSTATION_API_KEY}" }
    }
  }
}

More detail in Jev in Cursor.

VS Code (GitHub Copilot agent mode) — .vscode/mcp.json, with the key requested once and stored by VS Code instead of written to the file:

{
  "inputs": [
    {
      "type": "promptString",
      "id": "jevstation-key",
      "description": "JevStation API Key",
      "password": true
    }
  ],
  "servers": {
    "jevstation": {
      "type": "http",
      "url": "https://jevstation.com/api/mcp",
      "headers": { "Authorization": "Bearer ${input:jevstation-key}" }
    }
  }
}

Any other client — if it supports remote Streamable HTTP servers with custom headers, use the same URL and header. You can check the server by hand; tools/list costs nothing:

curl -s https://jevstation.com/api/mcp \
  -H "Authorization: Bearer $JEVSTATION_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Accept: application/json, text/event-stream" \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'

Billing

Tool calls draw on the same credit balance as the API and the playground. An evaluation costs 1 credit when the state is at most 8,000 characters and there are at most 5 questions, and 3 credits above either limit. initialize, tools/list and jev_list_question_sets are free. If an evaluation fails or returns no answers, the credits are refunded automatically. New accounts start with 200 free credits; packs are on the pricing page.

Troubleshooting

• 401 on connect. The key is missing, mistyped or deleted, or the header did not reach the server — often because the environment variable was not set when the client started. The server replies with HTTP 401 before any tool runs. Create a new key if needed and restart the client.
• Tool errors come back as results. Once authenticated, problems with a call are returned as a normal result with isError: true and a message the agent can read, not as a protocol error. Typical messages: bad or empty state, state too large, invalid questions, unknown question_set_id.
• Not enough credits (402 in the HTTP API). Over MCP this is an isError result that states the cost and your balance. Top up on pricing.
• Rate limited (429 in the HTTP API). Also an isError result. Wait before retrying and avoid tight loops; GET /api/v1/systemone with the same key shows your current quota.
• Evaluation failed. The result says the credits were refunded. Retrying is safe.
• Unknown tool. Returned as a JSON-RPC error. Refresh the tool list in your client.

Security notes

• The key spends credits. Anyone holding it can run evaluations on your balance. Treat it like a password.
• Keep it in an environment variable (JEVSTATION_API_KEY) or the client’s secret input. Never commit a config file that contains the literal key — in shared files use variable references such as ${JEVSTATION_API_KEY} or ${env:JEVSTATION_API_KEY}.
• Do not put secrets into tool arguments. state and tool_args are sent to the model for evaluation. When guarding a tool call, redact tokens, passwords and connection strings first.
• Delete leaked keys at Settings › API Keys and create a new one.
• Jev is a filter, not a security boundary. Keep allow-lists and permissions underneath jev_guard_tool_call; the harness walkthrough shows how the pieces fit.

References

Client syntax verified on 2026-09-28 against:

• Claude Code MCP: https://code.claude.com/docs/en/mcp
• Cursor MCP: https://cursor.com/docs/mcp
• VS Code MCP configuration reference: https://code.visualstudio.com/docs/agents/reference/mcp-configuration