mcp -ollama-deep-researcher

This MCP server implements a local, secure extension using the Model Context Protocol (MCP) over stdio. The Ollama Deep Researcher DXT Extension enables topic research by orchestrating web searches via Tavily, Perplexity, and Exa, combined with LLMs like Ollama or DeepSeek, all within a configurable research workflow. The server exposes standard MCP tools such as research, get_status, and configure. You can start a research session, poll its status, and adjust parameters like maxLoops, llmModel, and the search API to tailor results. Logs are emitted to stderr to aid debugging, and long-running subprocesses are guarded with timeouts to avoid hangs.

Prerequisites:

• Node.js and npm installed on your system
• Python (and pip) installed if you pursue Python-based tooling in the assistant logic
• Access keys for web search APIs (Tavily, Perplexity, Exa) if you intend to run web searches

Installation steps:

1. Clone the repository and install dependencies

git clone <your-repo-url> 
cd mcp-server-ollama-deep-researcher
npm install

2. Install Python dependencies for the assistant (in case you run the Python logic directly)

cd src/assistant
pip install -r requirements.txt
# or use your preferred Python environment manager

3. Set required environment variables for web search APIs (example)

export TAVILY_API_KEY=your_tavily_key
export PERPLEXITY_API_KEY=your_perplexity_key
export EXA_API_KEY=your_exa_key

4. Build the TypeScript server (if needed)

npm run build

5. Run the MCP server locally for testing

node dist/index.js
# Or run via your DXT host per its documentation

Notes:

• Ensure API keys are present in your environment when performing web searches; keys are not logged.
• The server uses MCP over stdio, so it should be integrated with a compatible DXT host.
• Research subprocesses are killed after 5 minutes to prevent hangs; adjust timeout settings if needed in your environment.
• Logs and errors are emitted to stderr to aid debugging; check logs when troubleshooting.
• If you modify configuration (maximizeLoops, llmModel, searchApi), use the configure tool via MCP to apply changes at runtime.