Apify MCP Deep Search

MCP Deep Search by Apify is a wrapper around MCP Deep Web Search provided by @just-every GitHub. It allows users to conduct complex research utilizing real time search across multiple search engines and models such as Google, Brave, Anthropic, OpenAI, Perplexity, and xAI. The MCP Deep Search Server now runs in a cloud environment and not locally on your computer.

Click to see how

You can limit the search to a single search engine or utilize multiple.

🔧 Tools

• deep_search
  Perform deep web searches to find current information, research topics, or answer questions using real-time data from multiple search providers. Use this when you need up-to-date information beyond your knowledge cutoff.

  • query - Your prompt / question
  • provider - select none or one from these providers: anthropic, brave, brave-images, openai, google, xai, sonar. If you don't select any, the tool shall select one to yield the best results. It is recommended though to specify it.

• comprehensive_research
  Perform in-depth research on complex topics using AI agents that automatically search multiple sources, analyze findings, and compile comprehensive reports. Ideal for thorough investigations, market research, technical analysis, or any topic requiring deep understanding from multiple perspectives.

  • query - Your prompt / task / question
  • modelClass
    • default - reasoning_mini
    • options - 'standard' | 'mini' | 'reasoning' | 'reasoning_mini' | 'monologue' | 'metacognition' | 'code' | 'writing' | 'summary' | 'vision' | 'vision_mini' | 'long' | 'image_generation' | 'embedding' | 'voice' | 'transcription';

⚠️ Caveat

Tool comprehensive_research runs LLM models and engines matching the provided API keys. If any of the API keys doesn't have enough credits, the task will fail and all the API calls will irreversibly burn your credits.

Organization verification for OpenAI

The streaming ability is conditioned by verification of the organization which API key is being used. If the organization is not verified, the comprehensive research will never finish successfully as long as you provide your OPENAI_API_KEY on the actor start.

This happens for modelClasses: 'reasoning_mini' and maybe some others. Try to set standard.

How to use it

Provide API keys and start the standby actor.

Search in the logs for the URL on which the server is exposed.

Copy this URL and build the final URL for your client:

HTTP STREAMABLE:

{
    "mcpServers": {
        "apify-deep-search": {
            "type": "mcp",
            "url": "<URL>/mcp",
        }
    }
}

SSE (legacy):

{
    "mcpServers": {
        "apify-deep-search": {
            "type": "sse",
            "url": "<URL>/sse",
        }
    }
}

How to start Actor in Standby mode

1. Provide input in the actor UI
2. Create a task by clicking the "Create Task" in the top-right corner
3. This should open the Task tab. If not, Tab > Standby Tasks > Your Task
4. Tab Standby
5. Click the "Test endpoint" button

If the request failed, check that the input contains all the required fields.

Environment Variables

• [Required] Google search API Key - makersuite.google.com
• [Optional] Brave search API Key - brave.com/search/api
• [Optional] Anthropic search API Key - console.anthropic.com
• [Optional] OpenAI search API Key - platform.openai.com
• [Optional] Perplexity searches (sonar, sonar-pro, sonar-deep-research) API Key - openrouter.ai
• [Optional] X.AI search API Key - x.ai

Chat with MCP using Inspector client

1. Install Inspector on your PC and start it

$npx @modelcontextprotocol/inspector

2. Click the link with token which opens the Inspector in the browser. If Proxy Session Token is missing (Configuration > Proxy Session Token), copy it from the terminal manually.
3. Set inspector to Streamable HTTP
4. Provide URL Example: <URL>/mcp
5. Set request timeout to 300000ms - Configuration > Request Timeout
6. Connect
7. Start chatting

Note: The standby server gets shut down after about 10 minutes of inactivity. If that happens, send a message to wake the server up and reconnect the Inspector client.

Billing

This actor doesn't charge for any calls. User only pays for the running server.

Credits

All credits go to the creators of the MCP Deep Search server.

GitHub
NPM