Deep Research MCP Server

Content

Deep Research MCP Server exposes a self-hostable deep research workflow to MCP clients. It can generate research plans, gather information from configured web search providers or local knowledge sources, and produce structured reports using separate thinking and task model settings.

The upstream project supports both Server-Sent Events and Model Context Protocol access. Its README documents Streamable HTTP at /api/mcp and SSE at /api/mcp/sse, plus MCP-specific environment variables for provider, search, thinking model, and task model configuration.

Source Review

• https://github.com/u14app/deep-research
• https://github.com/u14app/deep-research#model-context-protocol-mcp-server
• https://github.com/u14app/deep-research/blob/main/docs/deep-research-api-doc.md
• https://research.u14.app
• https://hub.docker.com/r/xiangfa/deep-research

These sources were reviewed on 2026-06-05. Prefer the live README and API docs for current deployment paths, MCP endpoints, environment variables, model provider options, search provider options, and transport support.

Features

• MCP support through Streamable HTTP and SSE transports.
• Deep research report generation with configurable thinking and task models.
• Search provider support for model-native search and external search services.
• Multi-LLM support across Gemini, OpenAI, Anthropic, DeepSeek, Grok, Mistral, Azure OpenAI, OpenRouter, Ollama, and OpenAI-compatible providers.
• Local knowledge-base workflows with uploaded text, Office, PDF, and other resource files.
• Research history, further research, report editing, translation, and knowledge graph features in the web app.
• Docker, Vercel, Cloudflare Pages, static export, and local development deployment paths.

Installation

Deploy Deep Research first, configure the MCP environment variables for model and search provider behavior, and then point your MCP client at the deployed MCP endpoint:

{
  "mcpServers": {
    "deep-research": {
      "url": "https://YOUR_DEEP_RESEARCH_DEPLOYMENT/api/mcp",
      "transportType": "streamable-http",
      "timeout": 600,
      "headers": {
        "Authorization": "Bearer YOUR_ACCESS_PASSWORD"
      }
    }
  }
}

The upstream README also documents an SSE MCP endpoint. Use a longer timeout than a simple lookup tool because deep research jobs can run for several minutes.

Use Cases

• Ask Claude to run a structured research workflow from an MCP client.
• Compare different thinking, task, and search providers for the same topic.
• Generate first-pass research briefs with citations for later human review.
• Use uploaded documents as a local knowledge source for research reports.
• Self-host a research service for teams that want explicit provider and deployment control.

Safety and Privacy

Deep Research can make many web search and model-provider calls from one agent request. Configure provider budgets, access controls, rate limits, and timeout behavior before exposing it to shared or autonomous workflows. Treat generated reports as drafts until citations and source quality are reviewed by a person.

Research prompts, uploaded files, search queries, local knowledge-base content, reports, citations, API keys, access passwords, and deployment logs can contain sensitive data. Review deployment hosting, model-provider, and search-provider retention policies before routing private or regulated information through the service.

Duplicate Check

No u14app/deep-research entry or matching source URL was found in content/mcp. Existing deep-research mentions in the repository are role or workflow references, not this self-hosted MCP research service.