# MCP

Agents that support remote MCP servers can call Search, Contents and Research as tools, with nothing to install. The server is /v1/mcp on the API host, over Streamable HTTP with your API key as a Bearer token.

## Command-line agents

Set AREX_API_KEY in your shell, then add the server with your agent's command. Claude Code and Kimi Code save the key into their configuration when the command runs.

### Codex

```bash
codex mcp add arex --url https://api.arex-research.com/v1/mcp \
  --bearer-token-env-var AREX_API_KEY
```

### Claude Code

```bash
claude mcp add --transport http arex https://api.arex-research.com/v1/mcp \
  --header "Authorization: Bearer $AREX_API_KEY"
```

### Kimi Code

```bash
kimi mcp add --transport http arex https://api.arex-research.com/v1/mcp \
  --header "Authorization: Bearer $AREX_API_KEY"
```

## Other clients

Clients configured with JSON use this entry, for example ZCode (Settings → MCP Servers → full configuration) or WorkBuddy (~/.workbuddy/mcp.json). Replace YOUR_API_KEY with your key.

### mcp.json

```json
{
  "mcpServers": {
    "arex": {
      "type": "http",
      "url": "https://api.arex-research.com/v1/mcp",
      "headers": { "Authorization": "Bearer YOUR_API_KEY" }
    }
  }
}
```

## Agent frameworks

Pi SDK, LangGraph and other frameworks can load the tools straight from the server.

### LangGraph

```python
import asyncio
import os
from langchain_mcp_adapters.client import MultiServerMCPClient
from langgraph.prebuilt import create_react_agent

async def main():
    client = MultiServerMCPClient({
        "arex": {
            "transport": "streamable_http",
            "url": "https://api.arex-research.com/v1/mcp",
            "headers": {"Authorization": f"Bearer {os.environ['AREX_API_KEY']}"},
        }
    })
    tools = await client.get_tools()
    agent = create_react_agent("openai:gpt-4.1", tools=tools)
    result = await agent.ainvoke({
        "messages": [{"role": "user", "content": "Find recent research on agent evaluation."}]
    })
    print(result["messages"][-1].content)

asyncio.run(main())
```

## Tools

Each tool takes the same fields as its API request.

| Tool | API operation |
| --- | --- |
| `arex_search` | `POST /v1/search` |
| `arex_contents` | `POST /v1/contents` |
| `arex_research` | `POST /v1/research` |

## Results and errors

A tool returns the API's JSON response. A failed call returns a tool error containing the API's error object. Rate limits are the same as the API's. arex_research waits until the Run ends.