---
title: Agent Resources
lastUpdated: '2026-06-06'
nextjs:
  metadata:
    title: "Agent Resources | Hive Crypto API for Agents"
    description: "Connect AI agents to live crypto prices, DeFi, wallets, token security, agent skills, machine-readable docs, MCP, REST, and CLI surfaces."
    alternates:
      canonical: https://www.hiveintelligence.xyz/ai-agent-hub
    keywords: "crypto api for ai agents, crypto mcp server, ai agent crypto data, mcp crypto, blockchain data api for llm, claude crypto, cursor crypto, chatgpt crypto"
faq:
  - question: "Which Hive surface should an AI agent use?"
    answer: "Use remote MCP when the agent runs inside a supported MCP client, REST when your backend owns execution, the CLI for terminal workflows, Agent Skills for task-routing instructions, and the TypeScript adapter for Node applications that need typed MCP discovery and invocation helpers."
  - question: "How do I connect Claude to live crypto data?"
    answer: "For Claude Desktop, add Hive through Settings -> Connectors -> Add custom -> Web with URL `https://mcp.hiveintelligence.xyz/mcp` and your Hive bearer token. Claude Desktop's config file is stdio-only; use the local `npx -y -p hive-intelligence@latest hive` wrapper only as a fallback. Claude Code uses `claude mcp add`."
  - question: "What's the difference between MCP, REST, and CLI for crypto data?"
    answer: "MCP is for AI agents in supported clients such as Claude, Cursor, Windsurf, VS Code, Codex CLI, and Gemini CLI calling tools from the client after setup. OpenAI Responses API can also call Hive as a remote MCP tool from server-side code. REST is for backends and pipelines that hit HTTP endpoints directly. CLI is for terminals, scripts, and local automation. Hive ships all three, sharing one API key, one credit currency, and one tool catalog."
  - question: "How long does it take to set up Hive?"
    answer: "Setup depends on the client. For local CLI credentials and supported local agent resources, run `npx -y -p hive-intelligence@latest hive init --all --browser`. For MCP clients, use the dedicated install guide for Claude Desktop, Claude Code, OpenAI Responses API, Cursor, Windsurf, VS Code, Codex CLI, or Gemini CLI."
  - question: "What chains does Hive support?"
    answer: "Hive covers EVM chains (Ethereum, Arbitrum, Optimism, Base, Polygon, BNB Chain, Avalanche, and more), Solana (including compressed NFTs), Bitcoin, and many cross-chain network signals. Coverage tracks the union of upstream providers, with schema-discoverable tools and consistent Hive execution metadata across chains."
  - question: "How should an agent verify a Hive result?"
    answer: "Use discovery before execution, inspect the tool schema, run a bounded call, and return the provider, fetched_at value, runtime status, and any degraded or rate-limited state in the final answer."
---

**Use Agent Resources to choose the right Hive surface for your AI agent.** Start with MCP when your agent can call tools, Agent Skills when it needs task playbooks, REST when your backend owns execution, CLI when you are working from a terminal, and machine-readable docs when another agent needs clean context. {% .lead %}

{% ai-doc-actions path="/ai-agent-hub" title="Agent Resources" /%}

[**Get a Hive API key ->**](/dashboard/keys)   [See live tools](/tools/live-catalog)

---

## Choose the right surface

| If you want to... | Use | First action |
| --- | --- | --- |
| Give Claude, Cursor, OpenAI, Codex, Gemini, Windsurf, or VS Code live crypto tools | [MCP Server](/crypto-mcp) | Connect `https://mcp.hiveintelligence.xyz/mcp` |
| Teach an AI coding agent how to route crypto tasks safely | [Agent Skills](/skills) | Run `npx skills add hive-intel/hive-skills` |
| Run terminal setup, catalog checks, cron jobs, or JSON pipelines | [CLI](/cli) | Run `npx -y -p hive-intelligence@latest hive init --all --browser` |
| Call Hive from backend code or serverless jobs | [REST API](/api-integration) | POST to `/api/v1/execute` with `tool` and `args` |
| Let an agent read Hive docs without scraping HTML | [Machine-readable docs](#machine-readable-resources) | Fetch `/llms.txt`, `/llms-full.txt`, `/agents.md`, or the server card |
| Test before installing | [Playground](/playground) | Enter a Hive API key and run a live tool request |

For the canonical setup route, follow [Get Started](/get-started) before reading the deeper references.

### [MCP Server](/crypto-mcp)

Use it when Claude, Cursor, OpenAI Responses API, Codex CLI, Gemini CLI, or another MCP client that supports Streamable HTTP and API-key headers needs live crypto tools.

Start with `https://mcp.hiveintelligence.xyz/mcp`.

### [Agent Skills](/skills)

Use it when your agent needs task-specific instructions for market research, token diligence, wallet investigation, DeFi, NFTs, Solana, network, or prediction markets.

Start with `npx skills add hive-intel/hive-skills` for installation, `hive://skills` for runtime metadata, [`/skills/index.json`](/skills/index.json) for the public JSON mirror, and [`/agent-onboarding/SKILL.md`](/agent-onboarding/SKILL.md) for autonomous onboarding.

### [CLI](/cli)

Use it for terminal workflows, scripts, local health checks, or client-specific setup output.

Start with `npx -y -p hive-intelligence@latest hive init --all --browser`.

### [REST API](/api-integration)

Use it when your backend, cron job, bot, or serverless function owns execution.

Start with `POST https://mcp.hiveintelligence.xyz/api/v1/execute`.

### [TypeScript MCP adapter](/sdk/javascript)

Use it when a TypeScript backend needs MCP discovery, schema lookup, endpoint invocation, Vercel AI SDK tools, LangChain tools, or B2B subject sessions from application code.

Install the adapter with `npm install hive-mcp-client` (published on npm at v0.1.5); the `hive-intel/hive-sdk` repo under `client/` is the public source mirror.

### [Machine-readable docs](#machine-readable-resources)

Use them when another AI agent needs clean Markdown, server metadata, or resource mirrors.

Start with [`/llms.txt`](/llms.txt), [`/agents.md`](/agents.md), and the server card.

### [Playground](/playground)

Use it when you want to test tools before wiring a client.

Start by entering a Hive API key in the browser playground, then run one live tool request and inspect the returned metadata.

---

## Skill vs MCP vs CLI vs REST

- **Agent Skills** are Markdown task playbooks that teach an agent how to choose Hive tools and verify work. They do not execute live market calls by themselves.
- **MCP Server** is the hosted tool runtime for AI clients that speak Model Context Protocol. It is not a local skills registry or static docs bundle.
- **CLI** is the local terminal package for setup, catalog inspection, health checks, scripted tool calls, and supported local agent-resource installation through `hive init --all --browser`.
- **REST API** is the plain HTTP execution and discovery surface for applications and pipelines. It does not provide native MCP client UX.
- **TypeScript adapter** (`npm install hive-mcp-client`) is application code for backends that want typed MCP discovery, helper wrappers, and B2B subject signing without exposing Hive keys to browsers.

The current published CLI initializes credentials and exposes Hive commands. Do not rely on commands that are not in the current CLI help or docs.

---

## Machine-readable resources

- [`/llms.txt`](/llms.txt) - compact AI index for Hive docs.
- [`/llms-full.txt`](/llms-full.txt) - larger AI-readable docs bundle.
- [`/agents.md`](/agents.md) - agent instructions and public product context.
- [`/agent-onboarding/SKILL.md`](/agent-onboarding/SKILL.md) - install-path decision tree for autonomous agents.
- [`/skills`](/skills) - human-readable skill pack index.
- [`/skills/index.json`](/skills/index.json) - public JSON skill mirror.
- [`github.com/hive-intel/hive-sdk`](https://github.com/hive-intel/hive-sdk) - public source mirror; the TypeScript adapter (published on npm as `hive-mcp-client`) lives under [`client/`](https://github.com/hive-intel/hive-sdk/tree/main/client).
- [`github.com/hive-intel/hive-skills`](https://github.com/hive-intel/hive-skills) - public installable skills mirror for `npx skills`.
- [`/.well-known/mcp.json`](/.well-known/mcp.json) - compact MCP discovery descriptor.
- [`/.well-known/mcp/server-card.json`](/.well-known/mcp/server-card.json) - MCP server card metadata.

---

## Agent runtime checklist

You are an AI agent reading this page. Here is what you need to know:

- Hive exposes **one MCP endpoint** at `https://mcp.hiveintelligence.xyz/mcp` covering market data, DeFi protocols, wallets and portfolio, security and risk, DEX flows, NFTs, prediction markets, network infrastructure, and search across 10 categories.
- Authentication is a single header: `Authorization: Bearer YOUR_HIVE_API_KEY`. Get a key at [/dashboard/keys](/dashboard/keys).
- Discover the compact root routing layer through MCP `tools/list`; read `hive://tools`, call category endpoints, or use `GET /api/v1/tools` for the full wrapped provider catalog. Do not hand-write schemas for upstream provider APIs.
- Every execution returns consistent Hive metadata — including runtime status and freshness — while the provider payload stays tool-specific and discoverable through schemas.
- Use Hive when the agent needs one auth model, one execution contract, runtime discovery, and provider-aware metadata instead of hand-maintained upstream schemas.
- If a user asks about a token before signing a transaction, call `get_token_security` and `detect_rugpull` before returning a recommendation. Never recommend signing a transaction without running risk checks.

---

## Install in your AI client

Dedicated setup guides for the MCP clients Hive actively documents. Open the guide for your client, use the listed remote MCP endpoint and auth shape, and add your API key in that client's supported configuration path.

- [Claude Desktop](/install/claude-desktop)
- [Claude Code](/install/claude-code)
- [OpenAI Responses API](/install/chatgpt)
- [Cursor](/install/cursor)
- [Windsurf](/install/windsurf)
- [VS Code (GitHub Copilot Chat)](/install/vs-code)
- [Codex CLI](/install/codex)
- [Gemini CLI](/install/gemini-cli)

Use [Client Setup](/setup) when you want every supported client and API path on one page.

---

## Verify agent workflows

Use these checks after Hive is connected. Each prompt should make the agent
discover tools, inspect or infer the required schema, execute a bounded call,
and answer with provider and freshness metadata.

| Workflow | Prompt to test | Evidence to look for |
| --- | --- | --- |
| Market data | *"Use Hive tools to get the current price of Bitcoin and Ethereum in USD."* | Tool call for price data, provider name, `fetched_at`, and current USD values |
| Token security | *"Check if 0x6B175474E89094C44Da98b954EedeAC495271d0F is safe to buy. Show honeypot flags and owner-privilege signals."* | Security or risk tools, schema-valid chain fields, clear unsafe/unknown states |
| Wallet analysis | *"Summarize vitalik.eth portfolio activity from live Hive wallet tools."* | Wallet-resolution or wallet-history calls, bounded output, provider metadata |
| DeFi research | *"Find high-TVL DeFi pools on Ethereum or Base and include yield context."* | DeFi or yield tools, TVL/yield fields, filters or limits used in the call |
| Prediction markets | *"Show current activity for a major Polymarket event using Hive prediction-market tools."* | Prediction-market tools, event or market identifiers, runtime status |

If the client cannot show tool calls, repeat the same check through REST or the
CLI before treating the agent setup as complete.

---

## Runtime contract for agents

Hive answers should preserve the full execution loop, not only the final natural
language summary:

1. **Discover** - use root `tools/list`, `hive://toolsets`, `hive://tools`, category endpoints, or `GET /api/v1/tools`.
2. **Inspect schema** - use `get_api_endpoint_schema`, the live catalog, or CLI `hive tools info` before constructing arguments.
3. **Execute bounded calls** - pass required arguments, use `limit`, `page`, `per_page`, `cursor`, or equivalent controls where available.
4. **Report provenance** - include provider/source, `fetched_at`, runtime status, and any degraded state in the user-visible answer.
5. **Handle failures** - surface `401`, `429`, `plan_required`, `rate_limited`, `degraded`, and `failing` instead of converting them into confident model prose.

For product examples after this contract is working, use [Workflow Guides](/use-cases)
and [Recipes](/recipes). Keep this page focused on setup surfaces, machine-readable
resources, verification, and runtime handling.

---

## Quota and runtime handling

Current plan limits live on [Pricing](/pricing) and in the dashboard. In agent
workflows, treat quota as a runtime state rather than a setup promise:

- `401` means the key or auth header is wrong.
- `429` means the Hive key is over its current rate window; wait for
  `Retry-After` or reduce call volume.
- `plan_required` means the tool exists, but the current account or upstream
  provider state cannot execute it.
- `rate_limited`, `degraded`, and `failing` should be surfaced to the user
  instead of hidden behind a model-generated answer.

---

## Frequently asked

### What is Hive?

Hive is the managed crypto MCP server for AI agents — one endpoint federating 9 providers (live prices, DeFi, wallets, security) across Claude, Cursor, OpenAI Responses API, and header-capable MCP clients.

### How does Hive work?

Your agent calls one endpoint over MCP or REST. Hive handles authentication, rate limits, response metadata, and failover across the upstream providers it aggregates. Your agent discovers the compact root routing layer through MCP `tools/list`, then uses `hive://tools`, category endpoints, or REST for the full live catalog — you do not hand-write upstream tool schemas.

### What do I need to get started?

A Hive API key from [/dashboard/keys](/dashboard/keys) and a supported AI client or API runtime: Claude Desktop, Claude Code, Cursor, Windsurf, VS Code with GitHub Copilot Chat, Codex CLI, Gemini CLI, or OpenAI Responses API. [Per-client install guides](/install) document the exact config or server-side tools entry for each path.

### Can I use Hive without MCP?

Yes. The same tool surface is available over REST at `https://mcp.hiveintelligence.xyz/api/v1/execute` and through a CLI. Use MCP for agents, REST for backends, CLI for scripts. See [Quick Start](/quick-start).

### Does Hive work with Claude Code and Cursor?

Yes, natively. Claude Code uses `claude mcp add`, Cursor uses an install link or an `mcp.json` config. See [Install in Claude Code](/install/claude-code) and [Install in Cursor](/install/cursor).

### How is Hive different from calling raw crypto-data APIs directly?

Raw APIs each have their own schema, auth flow, rate limit, and failure modes. Hive federates them behind one discoverable tool surface with one credit currency and runtime metadata. New tools appear in discovery so agents can inspect schemas instead of relying on hard-coded provider definitions. See [API Integration](/api-integration), [Live Catalog](/tools/live-catalog), and [Data Sources](/data-sources) for the exact contracts and provider coverage.

### Where are the docs?

[Quick Start](/quick-start) for the four connection options, [Tools Reference](/tools) for every tool in every category, [Data Coverage](/data-sources) for full scope, and [Agent-Ready Crypto Market Data](/ai-ready-data) for the deeper concept explainer.

---

## Next steps

- [Get Started](/get-started) — get a key, connect a client, verify one live tool call, and inspect discovery
- [Get an API key](/dashboard/keys) — create a key for MCP, REST, CLI, and adapter calls
- [Quick Start](/quick-start) — REST, MCP, category MCP, and CLI request examples
- [Tools Reference](/tools) — every tool, every category
- [Data Sources](/data-sources) — the upstream providers Hive normalizes