MCP Integration Documentation

Endpoints & Gateway Architecture

All MCP requests route through our edge gateway at https://api.wiserlab.ai. We support both scoped and unified streaming endpoints:

Endpoint	Protocol	Description
GET/POST /mcp/{server_slug}	Streamable HTTP	Scoped endpoint connecting directly to a specific MCP server (recommended for Cursor, Claude Desktop & Claude Code).
GET/POST /mcp	Streamable HTTP	Unified gateway aggregating all authorized MCP servers into a single multi-server session.
POST /mcp-rest/tools/call	HTTPS POST	Stateless REST execution for invoking a specific tool by name without maintaining a stream.
GET /mcp-rest/tools/list	HTTPS GET	OpenAI-compatible function definitions for all authorized tools in the catalog.

Active Server Endpoints

Server	Slug	Streamable HTTP URL
Smart Search Engine	`smart_search`	`https://api.wiserlab.ai/mcp/smart_search`
WiserLab Tools	`wiserlab`	`https://api.wiserlab.ai/mcp/wiserlab`

Authentication & Keys

Authentication uses your standard Wiser Lab API key (prefixed with wsl-...). Our gateway edge translates this token seamlessly. Both standard and LiteLLM headers are supported:

HTTP Authorization Headers

# Standard Bearer Auth:
Authorization: Bearer wsl-...

# Or LiteLLM Proxy Compatible Header:
x-litellm-api-key: Bearer wsl-...

User Integrations & Tool Credentials (GitHub, AWS, etc.)

Manage Integrations

Certain MCP servers (such as github_tools or cloud infrastructure tools) require personal authentication to access private repositories, resources, or files on your behalf. Wiser Lab supports two seamless options:

Option 1: Client Header Pass-Through

If you prefer not to store credentials in your account, add the custom header directly into your local IDE client config (Cursor, Windsurf, Claude Desktop):

"X-MCP-User-Token": "YOUR_TOKEN"

Option 2: Account Vault

Recommended

Connect your personal token (e.g. GitHub PAT or cloud access key) once in your Integrations Vault. Tokens are encrypted at rest with AES-128 and automatically attached when your IDE calls the tool.

Zero IDE Config

Cursor IDE Integration

To configure tools in Cursor, open Cursor Settings > Features > MCP or edit your project-level configuration file at .cursor/mcp.json:

.cursor/mcp.json

{
  "mcpServers": {
    "brave_search": {
      "url": "https://api.wiserlab.ai/mcp/brave_search",
      "headers": {
        "Authorization": "Bearer wsl-..."
      }
    },
    "github_tools": {
      "url": "https://api.wiserlab.ai/mcp/github_tools",
      "headers": {
        "Authorization": "Bearer wsl-..."
      }
    }
  }
}

Tip: After saving, Cursor will automatically connect and show green status indicators next to each server in the MCP settings panel.

Claude Desktop Integration

Edit your Claude Desktop configuration located at:

macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
Windows: %APPDATA%\Claude\claude_desktop_config.json
Linux: ~/.config/Claude/claude_desktop_config.json

claude_desktop_config.json

{
  "mcpServers": {
    "brave_search": {
      "url": "https://api.wiserlab.ai/mcp/brave_search",
      "headers": {
        "Authorization": "Bearer wsl-..."
      }
    }
  }
}

Claude Code CLI Integration

To configure tools with the Claude Code CLI, add any active server directly via the terminal:

Terminal CLI Command

claude mcp add brave_search https://api.wiserlab.ai/mcp/brave_search \
  -H "Authorization: Bearer wsl-..."

Tip: Alternatively, you can supply the LiteLLM-compatible header: -H "x-litellm-api-key: Bearer wsl-..."

Codex Integration

For Codex CLI and extensions, configure remote MCP servers in ~/.codex/config.toml:

~/.codex/config.toml

[mcp_servers.brave_search]
url = "https://api.wiserlab.ai/mcp/brave_search"
transport = "http"
headers = { "Authorization" = "Bearer wsl-..." }

Windsurf & Continue.dev

For Codeium Windsurf, place your configuration in ~/.codeium/windsurf/mcp_config.json:

~/.codeium/windsurf/mcp_config.json

{
  "mcpServers": {
    "brave_search": {
      "serverUrl": "https://api.wiserlab.ai/mcp/brave_search",
      "headers": {
        "Authorization": "Bearer wsl-..."
      }
    }
  }
}

For Continue.dev in VS Code / JetBrains, place your configuration in ~/.continue/config.json:

~/.continue/config.json

{
  "experimental": {
    "modelContextProtocolServers": [
      {
        "transport": {
          "type": "streamable-http",
          "url": "https://api.wiserlab.ai/mcp/brave_search",
          "requestOptions": {
            "headers": {
              "Authorization": "Bearer wsl-..."
            }
          }
        }
      }
    ]
  }
}

AGY (Antigravity) Integration

For Google Antigravity (AGY) CLI and IDE, configure your MCP servers in ~/.gemini/config/mcp_config.json or project workspace .agents/mcp_config.json:

~/.gemini/config/mcp_config.json

{
  "mcpServers": {
    "brave_search": {
      "serverUrl": "https://api.wiserlab.ai/mcp/brave_search",
      "headers": {
        "Authorization": "Bearer wsl-..."
      }
    }
  }
}

Tip: Tools are automatically discovered and registered into the agent's context upon conversation start.

OpenCode Integration

Configure OpenCode remote MCP servers in your project or global ~/.config/opencode/opencode.json:

opencode.json

{
  "$schema": "https://opencode.ai/config.json",
  "mcp": {
    "brave_search": {
      "type": "remote",
      "url": "https://api.wiserlab.ai/mcp/brave_search",
      "headers": {
        "Authorization": "Bearer wsl-..."
      },
      "enabled": true
    }
  }
}

OpenClaw Agent Integration

For OpenClaw autonomous agents, you can register tools via the terminal CLI command or by configuring the JSON file:

Method 1: Terminal CLI Command (openclaw mcp set)

Terminal CLI Command

openclaw mcp set brave_search '{"url":"https://api.wiserlab.ai/mcp/brave_search","transport":"streamable-http","headers":{"Authorization":"Bearer wsl-..."}}'

Method 2: Config File (~/.openclaw/config.json)

~/.openclaw/config.json

{
  "mcp": {
    "servers": {
      "brave_search": {
        "url": "https://api.wiserlab.ai/mcp/brave_search",
        "transport": "streamable-http",
        "headers": {
          "Authorization": "Bearer wsl-..."
        }
      }
    }
  }
}

Tip: You can verify connectivity anytime using openclaw mcp doctor brave_search --probe.

Hermes Agent Integration

For NousResearch Hermes Agent, you can connect MCP servers via the terminal CLI command or by configuring the YAML file:

Method 1: Terminal CLI Command (hermes mcp add)

Terminal CLI Command

hermes mcp add brave_search --url https://api.wiserlab.ai/mcp/brave_search

Method 2: Config File (~/.hermes/config.yaml)

~/.hermes/config.yaml

mcp_servers:
  brave_search:
    url: "https://api.wiserlab.ai/mcp/brave_search"
    transport: "streamable-http"
    headers:
      Authorization: "Bearer wsl-..."

Tip: Run /reload-mcp inside Hermes chat or test the server with hermes mcp test brave_search.

Python SDK & Async Streams

Using the official Model Context Protocol Python SDK (pip install mcp):

agent_tool_caller.py

import asyncio
from mcp.client.streamable_http import streamablehttp_client
from mcp import ClientSession

async def run():
    endpoint = "https://api.wiserlab.ai/mcp/brave_search"
    headers = {"Authorization": "Bearer wsl-..."}

    async with streamablehttp_client(endpoint, headers=headers) as (read_stream, write_stream):
        async with ClientSession(read_stream, write_stream) as session:
            await session.initialize()
            
            # 1. Discover capabilities
            tools = await session.list_tools()
            print("Tools available:", [t.name for t in tools.tools])

            # 2. Execute a tool
            result = await session.call_tool(
                "brave_web_search",
                arguments={"query": "Wiser Lab AI high performance platform"}
            )
            print("Tool Result:", result.content)

asyncio.run(run())

Node.js SDK & TypeScript Integration

Connect via the official Model Context Protocol TypeScript SDK (npm install @modelcontextprotocol/sdk):

mcp-client.mjs

import { Client } from "@modelcontextprotocol/sdk/client/index.js";
import { StreamableHTTPClientTransport } from "@modelcontextprotocol/sdk/client/streamableHttp.js";

const transport = new StreamableHTTPClientTransport(
  new URL("https://api.wiserlab.ai/mcp/brave_search"),
  {
    requestInit: {
      headers: {
        Authorization: "Bearer wsl-..."
      }
    }
  }
);

const client = new Client(
  { name: "wiserlab-node-agent", version: "1.0.0" },
  { capabilities: {} }
);

await client.connect(transport);

// 1. List available tools
const tools = await client.listTools();
console.log("Tools available:", tools.tools.map(t => t.name));

// 2. Call an MCP tool
const result = await client.callTool({
  name: "brave_web_search",
  arguments: { query: "Wiser Lab high performance AI cluster" }
});
console.log("Result:", result.content);

LLM Chat Completions with Automatic MCP Tools

When querying our OpenAI-compatible /v1/chat/completions endpoint, pass MCP servers directly in the tools array. The gateway automatically resolves tool schemas, queries the upstream MCP server during reasoning, and provides completed results to the model:

chat_with_mcp_tools.py (§12.5 Example A)

from openai import OpenAI

client = OpenAI(
    base_url="https://api.wiserlab.ai/v1",
    api_key="wsl-..."
)

response = client.chat.completions.create(
    model="gpt-4o",
    messages=[
        {"role": "user", "content": "What are the latest developments in quantum computing and TSLA stock price today?"}
    ],
    tools=[
        {
            "type": "mcp",
            "server_name": "brave_search"
        }
    ]
)

print(response.choices[0].message.content)

Direct REST Tool Invocation

If your backend or pipeline doesn't use the MCP streaming protocol, you can call any tool synchronously via standard HTTPS POST:

POST /mcp-rest/tools/call

curl -X POST https://api.wiserlab.ai/mcp-rest/tools/call \
  -H "Authorization: Bearer wsl-..." \
  -H "Content-Type: application/json" \
  -d '{
    "server_slug": "brave_search",
    "name": "brave_web_search",
    "arguments": {
      "query": "Quantum computing breakthroughs"
    }
  }'

Error Codes & Troubleshooting

Status	Error Message	Resolution
401 Unauthorized	`Invalid API key format`	Ensure your Authorization header starts with `Bearer wsl-...`.
429 Too Many Requests	`Budget has been exceeded!`	Your balance has fallen to $0.00. Add AI credits in your Billing dashboard to restore instant access.
404 Not Found	`MCP server not found or inactive`	Check the server slug in the URL against our Server Catalog.
502 Bad Gateway	`MCP upstream connection failed`	The third-party upstream service or database is momentarily unreachable. The gateway automatically retries.

Model Context Protocol (MCP)

MCP Guide

Endpoints & Gateway Architecture

Active Server Endpoints

Authentication & Keys

User Integrations & Tool Credentials (GitHub, AWS, etc.)

Option 1: Client Header Pass-Through

Option 2: Account Vault

Cursor IDE Integration

Claude Desktop Integration

Claude Code CLI Integration

Codex Integration

Windsurf & Continue.dev

AGY (Antigravity) Integration

OpenCode Integration

OpenClaw Agent Integration

Hermes Agent Integration

Python SDK & Async Streams

Node.js SDK & TypeScript Integration

LLM Chat Completions with Automatic MCP Tools

Direct REST Tool Invocation

Error Codes & Troubleshooting