Agent Hub

No need to open the exchange — say one sentence, and AI executes it directly in Claude, Cursor, or Codex on your Bitget account to trade stocks and crypto.

Runs as a local process on your device · API Key signed locally with HMAC-SHA256, never leaves your machine · MIT open source

GitHub · API Docs · Get API Key · Telegram

Table of Contents

Code
Bitget AgentHub
│
├── What it does
├── Choose your install path
│
├── Installation guides
│   ├── MCP Server    ← Claude Desktop / Cursor / Windsurf / Codex, OAuth one-click connect
│   ├── CLI · bgc     ← Claude Code / Codex CLI / OpenClaw
│   ├── Skill         ← Works with CLI, teaches AI when and how to call Bitget
│   ├── Signal        ← Real-time data + AI analysis, no account needed
│   ├── bitget-mcp-server ← US-stock, ETF, and analyst data, no account needed
│   └── SDK           ← For developers building custom integrations
│
├── Safety
└── FAQ

What it does

The frustrating part of using AI for trading isn't the strategy — it's the constant switching between AI and your exchange. Ask AI, then manually place the order. Check the chart, then switch back to chat. AgentHub removes that gap: say it once in your AI, and it handles the rest.

Choose Your Install Path

Select based on the AI tool you use:

Code
Please read 【https://www.bitget.com/support/articles/12560603894122】to complete Bitget Agentic account authorization, then run npx @bitget-ai/bitget-agent-installer upgrade-all --target all.

Prerequisite: Node.js ≥ 20 (run node -v to confirm)

Agentic Account

An Agentic account connects via OAuth and trades in a fund pool separate from your main account balance — a different connection method from manually configuring an API Key. Once authorized, MCP/CLI tools immediately get trading permission for that account, no key entry required.

Paste this directly into your AI — it will walk you through browser OAuth and install the matching tool:

Code
Please read 【https://www.bitget.com/support/articles/12560603894122】to complete Bitget Agentic account authorization and install the matching MCP.

MCP Server

The install steps below work for all supported tools. The core command npx -y @bitget-ai/bitget-agent-mcp never changes — the only difference is where you put the config file. Find your tool in the Step 1 table.

Let AI Configure It For You (Recommended)

Paste the following directly into your AI and it will complete the setup automatically:

Code
Please read 【https://www.bitget.com/support/articles/12560603894122】to complete Bitget Agentic account authorization and install the corresponding MCP.

About This Authorization

Funds must be moved from your main account to your Agentic account via Web before trading — your AI can't do that transfer, and can't withdraw either. Once credentials are saved, trading tools work immediately, no Agent restart needed; if the connection drops or credentials expire, a full OAuth run is the only fix.

Manual Setup

Step 1 — Open your AI tool's config file:

Step 2 — Add the following config. No API Key needed: the first time you use a trading tool after startup, the AI will automatically walk you through a browser OAuth authorization, and your credentials are issued by Bitget's server and saved locally:

Code
{
  "mcpServers": {
    "bitget": {
      "command": "npx",
      "args": ["-y", "@bitget-ai/bitget-agent-mcp"]
    }
  }
}

Step 3 — Fully quit and restart your AI tool.

Step 4 — Type What's the current BTC price? in your AI. Getting market data back confirms the basic setup worked. Ask something that needs trading permission next (e.g. check my account balance) and the AI will automatically open your browser to walk you through OAuth authorization — see "About This Authorization" above.

Adjust Mode As Needed

Append parameters to args to switch modes:

Example — read-only mode:

Code
"args": ["-y", "@bitget-ai/bitget-agent-mcp", "--read-only"]

Proxy Setup (If Needed)

If you need to route traffic through a VPN or proxy, add an env block to your MCP config (proxy networking only — unrelated to account authorization; fill in your own proxy address):

Code
"env": {
  "HTTPS_PROXY": "your-proxy-address, e.g. http://127.0.0.1:port",
  "NODE_USE_ENV_PROXY": "1"
}

What AI Can Do After Setup

Load all modules: "args": ["-y", "@bitget-ai/bitget-agent-mcp", "--modules", "all"]

Try These

Code
What's the current BTC price?
Show my USDT balance and open futures positions
Open a BTC long on paper trading, 10x leverage, 0.01 BTC
Check BTC price, then place a limit buy 2% below current price

CLI · bgc

The install steps below are universal for all supported tools. After running the install command, it auto-detects which tools you have and deploys to each — no need to repeat the process per tool.

Install

Paste into your terminal AI, or run directly in your terminal:

Code
Please run the following command to install the Bitget Agent Hub terminal tools (requires Node.js 20+):
npx @bitget-ai/bitget-agent-installer upgrade-all --target all
Once done, verify by running `bgc --version` and `bgc discover`, and show me the output.

Code
npx @bitget-ai/bitget-agent-installer upgrade-all --target all

This installs bgc CLI, the trading Skill, and the market analysis Skill, and deploys them to Claude Code / Codex / OpenClaw.

Set Up API Key

Want to skip manual keys via OAuth instead? The CLI doesn't support that yet — use the MCP Server path above for one-click OAuth connection. No API Key yet? Go to API Management → Create API Key → enable Read + Trade permissions. Save all three values: API Key, Secret Key, Passphrase. Start with a Demo Key and use --paper-trading before going live.

Code
export BITGET_API_KEY="your-api-key"
export BITGET_SECRET_KEY="your-secret-key"
export BITGET_PASSPHRASE="your-passphrase"

Verify Install

Code
bgc --version         # check version
bgc discover          # list all available operations

Full Tool List

bgc covers all operations through 12 intent verbs:

Common Commands

Code
# Check account balance
bgc account_overview --coin USDT

# Preview an order (--dry-run won't actually send it)
bgc order --action place --category SPOT --symbol BTCUSDT \
  --side buy --orderType market --qty 0.001 --dry-run

# View futures positions
bgc position --category linear

# Preview a fund transfer
bgc transfer_funds --action transfer --fromType spot --toType mix_usdt \
  --amount 100 --coin USDT --dry-run

Safety flags:

Proxy Setup

If you need to route traffic through a VPN or proxy:

Code
export HTTPS_PROXY="your-proxy-address"

Or override the API base URL:

Code
export BITGET_API_BASE_URL="https://api.bitget.com"

Try These

Code
What's the current BTC price?
Check my USDT balance and transfer 500 USDT to my futures account
Open a BTC long on paper trading, 10x leverage, 0.01 BTC
Check my BTC positions — if I have none, open a 10x long on paper trading

Signal (No Account Needed)

No Bitget account. No API Key. Installation deploys 5 Skill files locally and registers a remote public MCP data service (https://datahub.noxiaohao.com/mcp) as the data source — AI analysis is based on real-time data returned by this service, not generated from thin air.

Five data directions:

Install

Paste into your AI, or run directly in your terminal:

Code
Please run `npx @bitget-ai/bitget-signal --target all` (requires Node.js 20+),
install the Bitget market analysis Skills, then remind me to restart my AI tool.

Code
npx @bitget-ai/bitget-signal --target all

Restart your AI tool after installing, then ask away.

Code
pip install pandas numpy

bitget-mcp-server (New)

bitget-mcp-server is a separate, read-only data service covering US equities/ETF market data and company fundamentals — a different product from Signal (crypto macro/sentiment/technical/news data), with no shared data source or credentials. Like Signal, it needs no Bitget account or API Key.

Code
claude mcp add bitget-mcp-server --transport http https://agent.bitget.com/mcp

Skill

Skill is an instruction file installed into your AI tool that tells it when to call bgc, how to build the command, and how to prompt you for confirmation before any write operation. Without Skill, AI has the tool but doesn't know how to use it.

Skill is included automatically when you install via upgrade-all. To deploy separately:

Code
npx @bitget-ai/bitget-agent-skill --target all

Restart your AI tool after deploying.

SDK (For Developers)

Install

Code
npm install @bitget-ai/bitget-agent-sdk

Requirements: Node.js ≥ 20 · ESM only · zero runtime dependencies · TypeScript types included

Quick Start

Code
import { loadConfig, buildTools, BitgetRestClient, safeInvoke } from "@bitget-ai/bitget-agent-sdk";

// Start read-only — safe default
const config = loadConfig({ modules: "all", readOnly: true });
const client = new BitgetRestClient(config);
const tools = buildTools(config);
const ctx = { config, client };

// Query public market data — no API Key needed
const market = tools.find((t) => t.name === "market")!;
const res = await safeInvoke(market, { action: "tickers", category: "SPOT", symbol: "BTCUSDT" }, ctx);

if (res.ok) console.log(res.data);
else console.error(res.error);

Config Options

Runtime Discovery

No need to memorize operation lists — use discover at runtime:

Code
const discover = tools.find((t) => t.name === "discover")!;

await safeInvoke(discover, {}, ctx);                                     // list all domains and tool counts
await safeInvoke(discover, { domain: "trade" }, ctx);                    // tools in the trade domain
await safeInvoke(discover, { tool: "market", action: "tickers" }, ctx);  // exact contract for one action
await safeInvoke(discover, { search: "funding" }, ctx);                  // keyword search

Error Handling

Use safeInvoke — never throws:

Code
const res = await safeInvoke(tool, args, ctx);
if (res.ok) {
  // res.data
} else {
  // res.error — ready to use as an LLM tool-call error response
}

For fine-grained control, use typed errors:

Code
import { BitgetApiError, RateLimitError, ConfigError } from "@bitget-ai/bitget-agent-sdk";

try {
  await tool.handler(args, ctx);
} catch (err) {
  if (err instanceof RateLimitError) { /* back off and retry */ }
  else if (err instanceof BitgetApiError) { console.error(err.code, err.message); }
  else if (err instanceof ConfigError) { /* missing or invalid credentials */ }
}

Integration Testing

The SDK includes a built-in MockServer — no real API calls needed:

Code
import { MockServer } from "@bitget-ai/bitget-agent-sdk/testing";
import { loadConfig, BitgetRestClient } from "@bitget-ai/bitget-agent-sdk";

const mock = new MockServer();
await mock.start();

const config = loadConfig({ modules: "market", baseUrl: mock.baseUrl, apiKey: "test", secretKey: "test", passphrase: "test" });
const client = new BitgetRestClient(config);

// write your tests against client ...

await mock.stop();

More: agent-sdk

Safety

AgentHub is designed so your credentials never leave your machine: API Keys are read only from environment variables, all requests are signed locally with HMAC-SHA256 and sent directly to api.bitget.com — no middleware, no telemetry, no log uploads. Agentic Account credentials follow the same principle: they're stored locally by the MCP/SDK after OAuth and are never typed or pasted by the AI.

AgentHub has four built-in protection layers:

Start with paper trading, verify everything behaves as expected, then switch to live. When creating an API Key, follow least-privilege: don't enable withdrawals unless you need them.

FAQ

Q: My AI tool isn't in the list — can I still use it?

A: Any client that supports the MCP protocol can connect via the MCP Server path. Any terminal AI that supports external command calls can use the bgc CLI.

Q: Does querying market data require an API Key?

A: No. Only account operations (checking balance, placing orders, transferring funds) require a Key.

Q: Will my API Key be exposed to the AI?

A: No. Keys are read only from local environment variables — they never appear in the conversation context and are never uploaded to any server.

Q: What's the difference between a manual API Key and an Agentic Account?

A: A manual API Key is created in Web and configured by you directly, sharing your main account balance. An Agentic Account connects via OAuth into an isolated fund domain, with credentials managed automatically by the MCP/SDK after you approve the connection.

Q: Can an Agentic Account withdraw funds?

A: No. Withdrawals and main-account operations must always be done manually in Web.

Q: Is bitget-mcp-server the same as Signal?

A: No. They are separate MCP servers. Signal covers crypto macro/sentiment/technical/news data; bitget-mcp-server covers US-listed equities, ETFs, and related fundamentals/analyst data.

Q: Bitget tools aren't fully loading in Cursor?

A: Cursor has a total limit of ~40 MCP tools. Disable other MCPs, or add --modules market to only load the market module.

Q: Will AI ask for confirmation before placing an order?

A: Yes. Skill guides AI to show a [CAUTION] prompt before any write operation and waits for your confirmation. High-risk operations like withdrawals also require the --confirm flag.

Q: Is it free?

A: AgentHub itself is MIT open source and free. Trading fees follow your Bitget account tier.

Q: Where do I report issues?

A: GitHub Issues · For security vulnerabilities, email security@bitget.com — do not post publicly.

Links