MCP Server for Deep Research

The server provides a specialized prompt template designed for comprehensive research tasks. When using the server, select the deep-research prompt to trigger a structured workflow that includes:

1. Question Elaboration: Expanding your initial question and defining scope.
2. Subquestion Generation: Creating focused sub-topics for systematic coverage.
3. Web Search Integration: Using Claude's web search to find authoritative sources.
4. Content Analysis: Synthesizing findings and providing citations.
5. Report Generation: Producing a structured research report as an artifact.

To use the Deep Research MCP Server with Claude Desktop on macOS, follow these steps:

1. Download and install Claude Desktop.
2. Run the setup script in your terminal to configure the environment:

   python setup.py

3. Once installed, start researching by selecting the deep-research prompt template within Claude and providing your research question.

For developers looking to build and distribute the server, use the following uv commands:

1. Sync dependencies:

   uv sync

2. Build distributions (generates source and wheel files in the dist/ directory):

   uv build

3. Publish to PyPI:

   uv publish

To use the published version of the Deep Research MCP Server, add the following configuration to your claude_desktop_config.json file. This uses uvx to run the server directly:

"mcpServers": {
  "mcp-server-deep-research": {
    "command": "uvx",
    "args": [
      "mcp-server-deep-research"
    ]
  }
}

The ResearchProcessor manages the lifecycle of a research task. It maintains a research_data dictionary and a list of notes.

Key fields in research_data include:

• question: The initial research question.
• elaboration: The expanded scope and definitions.
• subquestions: A list of specific questions used to structure the research.
• search_results: Data gathered from web searches.
• extracted_content: Content parsed from search results.
• final_report: The synthesized final output.

If you are developing the server locally and have not published it, you must add it to your Claude Desktop configuration file.

Configuration Paths:

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

Add the server using the uv command, specifying the absolute path to your local repository directory:

"mcpServers": {
  "mcp-server-deep-research": {
    "command": "uv",
    "args": [
      "--directory",
      "/Users/username/repos/mcp-server-application/mcp-server-deep-research",
      "run",
      "mcp-server-deep-research"
    ]
  }
}

The package provides a main() function that serves as the primary entry point to start the MCP server. It uses asyncio to run the server's main loop. This is typically used when running the package as a script or via a command-line interface.

from mcp_server_deep_research import main

if __name__ == "__main__":
    main()

The server provides a specialized prompt designed to guide an AI through a structured research process. This prompt instructs the model to elaborate on a question, generate subquestions, perform web searches, analyze results, and finally produce a formatted research report.

To use this prompt, you must provide the required argument research_question.

{
  "name": "deep-research",
  "arguments": {
    "research_question": "Your specific research topic here"
  }
}

The server exposes two primary resources that allow you to inspect the state of an ongoing research process:

1. research://notes: A text/plain resource containing all research notes generated during the process, separated by newlines.
2. research://data: An application/json resource containing the structured research_data dictionary, which includes keys for question, elaboration, subquestions, search_results, extracted_content, and final_report.

### Available Resources
- `research://notes` (text/plain): Notes generated during the research process
- `research://data` (application/json): Structured data collected during the research process

The server uses specific enumerations to identify prompts and their arguments:

• DeepResearchPrompts.DEEP_RESEARCH: The identifier for the deep research prompt ("deep-research").
• PromptArgs.RESEARCH_QUESTION: The identifier for the required research question argument ("research_question").

class DeepResearchPrompts(str, Enum):
    DEEP_RESEARCH = "deep-research"

class PromptArgs(str, Enum):
    RESEARCH_QUESTION = "research_question"