Skip to main content

Bring Jev AI into your agent

Connect an agent that supports remote MCP over Streamable HTTP to classify, route and score with your iajev account.

Buy creditsManage credits
  1. 1

    Choose how to connect

    Use account authorization if your client supports OAuth, or create a dedicated API key.

  2. 2

    Add JEV AI to your agent

    Add the server URL or merge the configuration below into your client’s MCP settings. Reload the connection after saving.

  3. 3

    Verify with a free request

    Ask your agent to check your balance and list presets before running a paid decision.

Connect your AI tools

Use Jev AI from Codex, Cursor or Claude Code with your existing credits.

Choose one connection method

You only need one of these options to connect your agent.

Connect with your account

Add this MCP URL in your client, choose OAuth sign-in, then approve access in your browser. You can revoke the connection in Settings → MCP.

https://iajev.com/mcp

Choosing Option A here only displays instructions. Start authorization from your client; the client supplies the callback and security parameters.

Open authorization from your client

Codex
  1. In the ChatGPT desktop app (formerly Codex App), open Settings → Plugins → MCPs → Add → Add MCP Server.
  2. Name the server jev, enter the MCP server URL shown above, and save it. Hover over the jev row in the MCP server list and click Authenticate. Then sign in and complete account authorization in your browser.
  3. You may need to restart Codex before starting a new session. Then use the free verification prompt below to check tool access.

Alternatively, with the Codex CLI installed, run the commands below in order. If jev is already configured, skip the add command. The login command starts browser authorization. When switching from API Key to OAuth, first remove bearer_token_env_var and any Authorization header from the jev entry in ~/.codex/config.toml.

codex mcp add jev --url https://iajev.com/mcp
codex mcp login jev
codex mcp list
Cursor
  1. Open Cursor Settings → Customize → MCPs → New MCP Server → User.
  2. Merge the configuration below into ~/.cursor/mcp.json and save it. Keep your other MCP servers.
  3. Return to MCPs → jev. If needed, choose Reload, then click Authenticate when Needs Authentication appears.
{
  "mcpServers": {
    "jev": {
      "url": "https://iajev.com/mcp"
    }
  }
}
Claude Code
  1. Run the command below in your terminal to add jev to your user-level MCP configuration.
  2. Start Claude Code with claude, enter /mcp, select jev, then choose Authenticate.
claude mcp add --transport http --scope user jev https://iajev.com/mcp

Your client opens the authorization page in your browser. Sign in if needed, review the application and requested permissions, then choose Allow access. Return to your client and check that jev is connected and its tools are available. Use the free verification prompt below to check the connection.

Running decisions uses your existing account credits. Connecting and querying do not cost iajev credits.

Check your connection

Send this prompt to your agent. A successful connection returns your account balance and available presets without spending credits.

Use jev to call get_credit_balance and list_presets. Show my available credits and the available presets. Do not call run_decision yet.

What your agent can do

get_credit_balance
Check available credits and expiry information.
Free
list_presets
Find ready-to-use decision definitions.
Free
run_decision
Classify, judge or score your context with explicit criteria.
Uses credits
get_decision_result
Retrieve a decision using its original idempotency_key.
Free

If a decision times out, ask your agent to look up the result with the original idempotency_key. Reuse that key when retrying to avoid paying for a second decision.

Try run_decision

Copy a prompt into your agent after verifying the connection. Running a decision uses account credits; copying a prompt does not. Each new decision needs a fresh idempotency_key. If the response is lost, query the result with that key before retrying with the same key.

Route a support ticket

Choose a team and compare the probabilities of each option.

Use jev to call run_decision once. Set state.message to "I was charged twice for my subscription this month. Please help me get the duplicate payment refunded." Set decision.type to "choice", decision.question to "Which team should handle this ticket?", and decision.choices to ["Billing", "Technical support", "Sales", "Account security"]. Generate a fresh UUID for idempotency_key and retain it for result lookup or retries. Show the selected team, confidence, candidate probabilities and credits used. Do not issue a refund or send the ticket.

Score product feedback

Evaluate feedback against explicit product criteria.

Use jev to call run_decision once. Set state.message to "Our team exports 200 reports every week. We need scheduled CSV exports because manual downloads take two hours and sometimes miss reports." Set decision.type to "score", decision.question to "How strongly does this feedback support prioritizing scheduled exports?", and decision.criteria to ["The problem is specific", "The task occurs frequently", "The time cost is measurable", "The requested outcome is clear"]. Generate a fresh UUID for idempotency_key and retain it for result lookup or retries. Show the returned score, criterion results and credits used. Do not invent a score range or create a roadmap item.

Having trouble connecting?

If tools are missing, reload the MCP connection and check the permissions you approved. If authentication fails, sign in again or check that your key is active and IAJEV_API_KEY is available to the client process. The server URL is an MCP endpoint, not a page to open in your browser.