Multi-Agent Deep Researcher

A local, open-source alternative to ChatGPT Deep Research. It takes a user's research query, performs deep web research, verifies and synthesizes findings, and returns a cited, structured report — accessible directly inside Cursor via MCP.

Prerequisites

1. Ollama: Must be installed and running locally.

2. DeepSeek-R1 (7B): Pull the model in Ollama: ollama run deepseek-r1:7b (or just deepseek-r1).

3. Serper.dev API Key: Get a free API key from serper.dev.

4. uv: Used for dependency management.

Related MCP server: MyWebSearch MCP

Setup

1. Install dependencies:

   uv sync

2. Configure environment: Edit .env and set your SERPER_API_KEY:

   SERPER_API_KEY=your_api_key_here

Cursor Integration

1. Go to Cursor Settings -> Features -> MCP Servers.

2. Click + Add new MCP server.

3. Name: crew_research

4. Type: command

5. Command: uv

6. Arguments: --directory <path_to_this_project> run server.py (Replace <path_to_this_project> with the actual absolute path to this folder).

7. Ensure the server starts successfully and shows "Running".

Now you can ask Cursor's AI (e.g. Composer or Chat) to perform deep research on a topic, and it will invoke the crew_research tool!

Streamlit Web App

If you prefer a visual interface, you can use the built-in Streamlit web app:

1. Run the app:

   uv run streamlit run app.py

2. Open your browser to http://localhost:8501.

3. Type your research query in the text box and click "Start Research"!

Architecture

• Web Search Agent: Searches the web using Google Search via Serper.dev.

• Research Analyst Agent: Deduplicates, structures, and verifies facts.

• Technical Writer Agent: Formats a polished markdown report with citations.