Connect your AI · Agent Setup Doctor

Give the AI agent you already use a reliable check

Choose your AI, choose a live task, get the exact configuration, check it, test the connection, and go to your first real result. A few minutes, no account until you want a key, and nothing is executed by this page.

Step 1 of 6 · Choose your AI. No account until you want a key; nothing is executed by this page.

Which AI agent do you already use?

It stays the brain; Jithox only gives it a reliable check. You chose Your Jithox connection; it is kept for the next step.

ClaudeCERTIFIEDClaude Desktop and Claude Code speak MCP over Streamable HTTP with OAuth.

ChatGPT / CodexCONFIG_READYChatGPT connectors and Codex speak MCP; the client completes OAuth on the Jithox consent screen.

GeminiCONFIG_READYGemini CLI reads an mcpServers block with the endpoint URL.

Grok / CursorCONFIG_READYCursor reads mcp.json and sends the access key on each call; Grok-based clients that speak MCP use the same endpoint.

Any MCP clientCONFIG_READYAny client that speaks MCP over Streamable HTTP needs the endpoint and nothing else.

Any HTTP or CLI agentCERTIFIEDOne plain HTTPS request with your key; curl is enough.

What this is, and is not

Your agent stays the brain. Jithox adds verified checks, never another chat.
Read-only today: every live tool looks things up in an official source and changes nothing.
What a tool may do is stated on its card; a level that is not proven is not claimed.
Every accepted call answers with a receipt you can verify without asking us.

Execution is not enabled

No Jithox capability sends, books, pays or changes anything today. The ladder from VERIFY to EXECUTE ONCE and RECEIPT AND RECOVERY is stated on the tools page; a level appears there only once it is implemented and proven.

What Jithox does today, and does not claim

Every live tool reads and changes nothing. The ladder from reading to acting has six levels; only the first is live, and a level appears only once it is proven.

Developer details: the six action levels

Level 1 · VERIFY Read-only checks from authoritative sources. Live.
Level 2 · COMPARE Compare options with explicit constraints and costs. Not claimed.
Level 3 · PREPARE Prepare an exact, bounded external action. Not claimed.
Level 4 · APPROVE Bind owner approval to an actionHash, amount, destination and expiry. Not claimed.
Level 5 · EXECUTE ONCE Perform the approved action exactly once through an official connector. Not claimed.
Level 6 · RECEIPT AND RECOVERY Reconcile outcome and cost, provide the receipt, recover safely. Not claimed.

Advanced: design a multi-bot plan with these tools in the Team Builder (Labs).