# Connect an AI Agent to Bitpont | MCP, Codex and OpenAI

Connect to the hosted MCP endpoint or run the downloadable server locally. Choose a product profile, check access and make a first discovery request.

Canonical: https://www.bitpont.com/developers/agent-connections/


## Choose a connection

Use https://www.bitpont.com/api/v1/mcp with a remote MCP client. For a local stdio client, [download the server](/assets/sdk/bitpont-agent-mcp.mjs) and launch it with Node.js 20 or newer. Standard trading and discovery work without registration. Store your private workspace on the host for recovery after restarts.

Server-side JavaScript
```
import {BitpontClient} from './bitpont-client.mjs';
const client = new BitpontClient({publicAccess:true});
const info = await client.agent('describeTool', {tool:'viewBridgeAssets'});
console.log(info.data.inputSchema);
```

## Load one product or the full catalogue

For a treasury assistant, add ?product=treasury to the remote URL. For stdio, pass --product treasury. Product profiles narrow the tools presented to the model. API scopes control access. Call tools/list until the response has no nextCursor.

## Connect Codex or the Responses API

For Codex, run codex mcp add bitpont --url https://www.bitpont.com/api/v1/mcp. The [starter kit](/help/ai-agents/#agent-downloads) includes a Responses API tool configuration with an explicit allowlist and approval setting. For local function calling, select definitions from the [OpenAI function catalogue](/assets/docs/bitpont-openai-tools.json) and dispatch by tool name using the tool directory.

## Verify the connection

Initialize the client and list tools, prompts and resources. Read bitpont://products/treasury for the product’s handoff rules. Open the operations-move prompt for an example on Base. The prompts and product resources bundled with the local download also work offline.

[Run a public example in the browser](/help/ai-agents/#agent-try) or [integrate Bitpont into an AI product](/business/ai-platforms/).


