Use FreeJev with your Agent

Connect remote MCP, use the CLI, or install the skill. Every entry point uses the same account and product API.

Remote MCP

Add this URL to your client, then sign in and review the permissions. No local MCP server is required.

https://freejev.org/mcp

Codex CLI

codex mcp add freejev --url https://freejev.org/mcp
codex mcp login freejev

Grok Build

grok mcp add --transport http freejev https://freejev.org/mcp

Open Grok Build and authorize when prompted. You can also open /mcps and use its sign-in action.

Start with a read-only usage action before submitting work. Review or revoke connected Agents.

CLI

Requires Node.js 22 or later. Create a project API key in API keys, then set FREEJEV_API_KEY through your secret store. Keep credentials out of request files and command arguments.

npm install -g freejev-client
freejev --help

freejev decide --request request.json — Run Jev judgments on text or JSON. Consumes account credits; never retries automatically.
freejev usage — Read remaining credits and recent usage metadata without running the model.

Successful output is JSON; errors go to stderr with a nonzero exit code. Use --request-id to identify one intended call. Billable requests are never retried automatically.

Existing local MCP configurations can continue using freejev mcp with the API key in the client's secret environment.

npm install -g freejev-client@latest
npm uninstall -g freejev-client

Skill

The skill teaches an Agent when and how to use these actions. It contains public instructions and fictional examples. It does not contain credentials or grant access.

freejev skill install --target codex
# Or for Grok Build:
freejev skill install --target grok

npx freejev-client skill install --target codex

The installer uses the shared ~/.agents/skills/freejev/ directory. Restart your Agent session to discover the skill. Connect remote MCP or configure the CLI key separately.

Download the public SKILL.md for manual installation. Re-running the installer updates an unmodified managed skill; edited files are protected.

freejev skill uninstall --target codex

Available actions

`jev_decide`

Run Jev judgments on text or JSON. Consumes account credits; never retries automatically.

Input schema

{
  "type": "object",
  "required": [
    "state",
    "questions",
    "request_id"
  ],
  "properties": {
    "state": {
      "oneOf": [
        {
          "type": "string"
        },
        {
          "type": "object"
        },
        {
          "type": "array"
        }
      ]
    },
    "questions": {
      "type": "object",
      "additionalProperties": {
        "type": "object",
        "required": [
          "type",
          "instructions"
        ],
        "properties": {
          "type": {
            "type": "string",
            "enum": [
              "choice",
              "noul",
              "score"
            ]
          },
          "instructions": {
            "type": "string"
          },
          "criteria": {
            "oneOf": [
              {
                "type": "object",
                "additionalProperties": {
                  "type": "string"
                }
              },
              {
                "type": "array",
                "items": {
                  "type": "string"
                }
              }
            ]
          }
        }
      }
    },
    "request_id": {
      "type": "string",
      "pattern": "^[-\\w:.]{1,120}$",
      "description": "Stable ID for one intended call. Reuse prevents another charge; results are not stored."
    }
  },
  "additionalProperties": false
}

`freejev_usage`

Read remaining credits and recent usage metadata without running the model.

Input schema

{
  "type": "object",
  "properties": {},
  "additionalProperties": false
}

Usage and recovery

1 credit = 1,000 actual input tokens. All channels share your account credits. Output tokens are free.

Authentication expired: reconnect or check your API key.
Insufficient quota: stop and review your balance. Connections cannot buy credits automatically.
Connection lost: check usage or the returned task ID before starting a new billable request.
Async actions: keep the returned task ID and query its status. A dropped connection does not cancel the task.

Task support depends on the published actions above. Uses your FreeJev account credits. Inputs are sent to FreeJev and its inference providers. Do not send secrets. Never retry billable actions automatically.

Data handling