---
title: Connect Hive MCP
lastUpdated: '2026-06-22'
nextjs:
  metadata:
    title: Connect Hive MCP - Get Started
    description: "Connect Hive MCP: create an API key, add one MCP client, verify a live tool call, inspect discovery, and choose the right production surface."
    alternates:
      canonical: https://www.hiveintelligence.xyz/get-started
    keywords: "hive mcp setup, get started with hive, crypto mcp quickstart, run first mcp call, hive quick start"
---

**Connect Hive MCP in one supported client and verify one live crypto answer.** Start with an API key, add the remote MCP server, make one provider-backed tool call, inspect discovery, and choose the right surface for your app or agent. {% .lead %}

{% ai-doc-actions path="/get-started" title="Connect Hive MCP" /%}

[**Get an API key ->**](/dashboard/keys)   [Open client setup](/setup)   [View live catalog](/tools/live-catalog)

---

## Connection sequence

| Step | Do this | You are done when |
| --- | --- | --- |
| 1 | Create a Hive API key | You have a `hive_live_...` key from [/dashboard/keys](/dashboard/keys) |
| 2 | Connect one AI client | Claude, Cursor, OpenAI Responses API, Codex, Gemini, Windsurf, or VS Code can see Hive |
| 3 | Verify one live tool call | The client answers from Hive instead of model memory |
| 4 | Verify tool discovery | The client can list or search Hive tools before execution |
| 5 | Pick the production surface | MCP, REST, CLI, Skills, or the TypeScript adapter is clearly assigned |

After these steps, Hive is connected in one client, live data calls work, and the tool-discovery loop is visible before you move into deeper references.

---

## 1. Create your key

Create or select a Hive API key at [/dashboard/keys](/dashboard/keys). Every plan can access the full tool catalog; your plan controls monthly credits and rate limits.

Use this header in new integrations:

```text
Authorization: Bearer YOUR_HIVE_API_KEY
```

The legacy `x-api-key` header is still accepted, but new docs and examples use Bearer auth.

---

## 2. Connect one AI client

Use the client you already work in. Switching clients to try Hive adds avoidable setup work.

Every client guide uses the same managed remote MCP server:

| Item | Value |
| --- | --- |
| Remote MCP URL | `https://mcp.hiveintelligence.xyz/mcp` |
| Transport | Streamable HTTP |
| Auth header | `Authorization: Bearer YOUR_HIVE_API_KEY` |

| Client or runtime | Start here | Verify by |
| --- | --- | --- |
| Claude Desktop | [Install in Claude Desktop](/install/claude-desktop) | Ask for BTC, ETH, and SOL prices |
| Claude Code | [Install in Claude Code](/install/claude-code) | Run `/mcp`, then ask for market data |
| Cursor | [Install in Cursor](/install/cursor) | Confirm Hive appears under Tools & MCP |
| OpenAI Responses API or ChatGPT | [Install in OpenAI / ChatGPT](/install/chatgpt) | Send a server-side remote MCP request |
| Codex CLI | [Install in Codex CLI](/install/codex) | Run `codex mcp list` or ask a crypto query |
| Gemini CLI | [Install in Gemini CLI](/install/gemini-cli) | Ask for a live price with Hive connected |
| Windsurf | [Install in Windsurf](/install/windsurf) | Confirm the MCP server is enabled |
| VS Code | [Install in VS Code](/install/vs-code) | Use Copilot Chat with Hive enabled |

For all supported client setup examples on one page, use [Client Setup](/setup).

## 3. Verify one live tool call

After the client is connected, start a fresh session and ask:

> *"Use Hive tools to get the current price of Bitcoin and Ethereum in USD."*

You are ready to continue when the answer includes a visible Hive tool call and live provider-backed data. If the client does not show a tool call, reload the client, re-check the API key, and confirm the MCP server is enabled.

If you are building a backend or want a client-independent smoke test, run the same first call over REST:

```bash
curl -X POST https://mcp.hiveintelligence.xyz/api/v1/execute \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer YOUR_HIVE_API_KEY" \
  -d '{
    "tool": "get_price",
    "args": {
      "ids": "bitcoin,ethereum",
      "vs_currencies": "usd"
    }
  }'
```

For the REST smoke test, you are ready to continue when:

- `ok` is `true`.
- The response includes current price data.
- `meta` includes provider, runtime status, and `fetched_at`.
- `401` errors are resolved by fixing the key.
- `429` errors are handled by waiting for the `Retry-After` window or checking usage in the dashboard.
- Tool-name errors are resolved through [Live Catalog](/tools/live-catalog) or authenticated `GET /api/v1/tools`.

---

## 4. Verify tool discovery

Hive's root MCP endpoint intentionally keeps `tools/list` compact. General-purpose agents should discover task tools, inspect schemas, then execute.

For REST discovery:

```bash
curl -X GET "https://mcp.hiveintelligence.xyz/api/v1/tools?search=wallet%20risk&limit=5" \
  -H "Authorization: Bearer YOUR_HIVE_API_KEY"
```

For root MCP workflows, the expected loop is:

1. Call `search_tools` or read `hive://toolsets`.
2. Call `get_api_endpoint_schema` for the exact tool.
3. Call `invoke_api_endpoint` with schema-valid `args`.
4. Report provider, freshness, runtime status, and any degraded state.

Use category-scoped MCP endpoints only when you intentionally want a smaller direct tool list for one domain such as market data, wallet analysis, security, or prediction markets.

---

## 5. Pick the production surface

| If you need | Use | Why |
| --- | --- | --- |
| Agent-native live tool calls | [Root MCP](/crypto-mcp) | Use for Claude, Cursor, OpenAI Responses API, Codex, Gemini, Windsurf, and VS Code |
| Backend execution or serverless jobs | [REST API](/api-integration) | Plain HTTP with the same `tool` and `args` execution contract |
| Terminal setup and scripts | [CLI](/cli) | Local catalog inspection, health checks, setup helpers, and JSON pipelines |
| Agent playbooks | [Agent Skills](/skills) | Teaches agents how to choose tools, inspect schemas, and report provenance |
| TypeScript application integration | [TypeScript adapter](/sdk/javascript) | Typed MCP discovery, endpoint invocation, Vercel AI SDK tools, and LangChain helpers |
| Focused agent with a smaller tool surface | [Category endpoints](/tools) | Direct `tools/list` for one category |

Before production rollout, assign one owner for each component: key storage, execution surface, tool discovery, error handling, rate-limit handling, and user-visible provenance.

---

## Production readiness

Before shipping:

- Store Hive keys in a secret manager or environment variable, never in client-side code.
- Verify the official MCP endpoint before approving a client connection: `https://mcp.hiveintelligence.xyz/mcp`. Do not approve a look-alike MCP URL or marketplace entry that points somewhere else.
- Decide whether your app will call root MCP, category MCP, REST, or the TypeScript adapter.
- Use client tool-call approval or review prompts where your MCP client supports them.
- Use live discovery before hard-coding tool names.
- Show `fetched_at`, source/provider, and runtime status when answering users.
- Handle `401`, `429`, provider unavailability, and degraded runtime status without hiding failures.
- Keep outputs bounded with `limit`, `cursor`, `page`, `per_page`, or equivalent arguments.
- Use token security and risk tools before any workflow that could lead a user to sign or swap.
- Review [MCP Security](/mcp-security) before enabling stateful tools, B2B subject writes, or downstream actions.

---

## Next steps

- [Quick Start](/quick-start) - fuller examples for REST, MCP, category MCP, and CLI
- [Install Hive](/install) - all supported AI clients
- [Agent Resources](/ai-agent-hub) - MCP, Skills, CLI, REST, adapters, and machine-readable docs
- [Live Catalog](/tools/live-catalog) - discovery surfaces and current catalog shape
- [Errors](/errors) - retry and failure handling