LDR's MCP server is the project's own official integration, exposing its multi-engine search, private-document analysis, and cited report generation through 8 tools over local STDIO transport. It's compelling for privacy-focused users who already run local or cloud LLMs and search infrastructure, but the docs are explicit that the server lacks authentication and rate limiting and is meant for local, single-machine use rather than network deployment.
Local Deep Research (LDR) is an open-source AI research assistant that runs local LLMs (Ollama, LM Studio, llama.cpp) or cloud models (OpenAI, Anthropic, Google, and more), searches across 10+ engines (arXiv, PubMed, Semantic Scholar, SearXNG, GitHub, etc.) plus a user's own private documents, and synthesizes findings into cited summaries or full reports. User data lives in a per-user, AES-256-encrypted SQLCipher database, with passwords never stored. The same repository ships an MCP server so Claude Desktop and Claude Code can drive these research capabilities directly; the docs state it is designed strictly for local use via STDIO transport, with no built-in authentication or rate limiting.
1) Install the package with MCP extras: pip install "local-deep-research[mcp]"; 2) set up an LLM source (local Ollama/LM Studio/llama.cpp, or a cloud provider API key for OpenAI/Anthropic/Google/etc.); 3) add an mcpServers entry to claude_desktop_config.json (Claude Desktop) or .mcp.json / ~/.claude/mcp.json (Claude Code) with command set to ldr-mcp and any needed LDR_* variables passed via env (e.g. LDR_LLM_PROVIDER, LDR_LLM_OLLAMA_URL, LDR_LLM_OPENAI_API_KEY); 4) restart the Claude client to pick up the config; for web search, also run SearXNG or another configured search engine.
{
"mcpServers": {
"local-deep-research": {
"command": "ldr-mcp",
"env": {
"LDR_LLM_PROVIDER": "openai",
"LDR_LLM_OPENAI_API_KEY": "sk-..."
}
}
}
}