Gigaxity Deep Research MCP
A deep research MCP server: multi-source search, result fusion and cited synthesis through a Qwen3 model on OpenRouter
- 66 ★ on the repository
- Last commit
- MIT
Extended access
Why this level
- Queries and found snippets go to search providers and the synthesis model, and OpenRouter and search service keys are kept in the environment
- Installing from source pulls Python dependencies and runs a local server
Install
- Works in
Manual install
{"mcpServers": {"gigaxity-deep-research": {"type": "stdio", "command": "/path/to/gigaxity-deep-research/.venv/bin/python", "args": ["/path/to/gigaxity-deep-research/run_mcp.py"], "env": {"RESEARCH_LLM_API_BASE": "https://openrouter.ai/api/v1", "RESEARCH_LLM_API_KEY": "YOUR_OPENROUTER_API_KEY", "RESEARCH_LLM_MODEL": "qwen/qwen3-30b-a3b-thinking-2507"}}}}Before this: `git clone`, a virtual environment, `pip install -e .` and `.env` from `.env.example`. Use your own paths.
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- Queries and found snippets go to search providers and the synthesis model, and OpenRouter and search service keys are kept in the environment
- Installing from source pulls Python dependencies and runs a local server
This is third-party code. Review the repository files before installing.
What it does
Gigaxity Deep Research combines parallel search across SearXNG, Tavily, LinkUp, Brave and Parallel, RRF fusion and synthesis with citation binding and contradiction detection. All of it is exposed as seven MCP tools: three primitives (`search`, `vertical_search`, `research`) and four deep-research tools (`ask`, `discover`, `synthesize`, `reason`). The same pipeline is available over a FastAPI REST API. The `search` tool gathers raw multi-source results without a model call. `vertical_search` takes one SearXNG category (videos, science, IT, docs, packages), `discover` widens the source set and `synthesize` folds findings into a cited answer. The recommended synthesis model is Qwen3-30B-A3B-Thinking through OpenRouter, but any OpenAI-compatible endpoint works, such as DeepSeek-R1 or a local vLLM. The server runs as an MCP stdio server through `run_mcp.py`, as a REST API or both. The `CLAUDE.md` and `AGENTS.md` files carry instructions for wiring the full stack into Claude Code, Codex, Cursor and Hermes. Search services come from separate providers, and each is priced or limited its own way.
Who it is for. Developers and analysts who want their own deep research MCP with citations and a choice of synthesis model.
Good fit when
- A question falls past the model training date and the agent needs fresh sources with citations
- Only raw collection from several search engines without a model call is needed: `search`
- Videos, science, IT, docs or packages are wanted in a separate result through `vertical_search`
- The synthesis model must run on your own hardware through vLLM, SGLang or llama.cpp
Not a fit when
- There are no search service keys and no OpenRouter key: synthesis and some sources will not work without them
- A ready cloud service without deploying Python and Docker is needed
- The research must cover local documents only
Example request
Research which LLM quantization approaches appeared in recent months, gather sources with `discover` and produce a cited answer with `synthesize`MCP
- Transport
- stdio
- Authentication
- API key
Environment variables
RESEARCH_LLM_API_KEYrequired, secret- OpenRouter or other OpenAI-compatible endpoint key for synthesis
RESEARCH_LLM_API_BASE- Model endpoint address, OpenRouter by default
RESEARCH_LLM_MODEL- Synthesis model, qwen/qwen3-30b-a3b-thinking-2507 recommended
RESEARCH_SEARXNG_HOST- Address of the SearXNG instance, the primary search source
Limitations
Synthesis needs an OpenRouter key (`RESEARCH_LLM_API_KEY`) or your own OpenAI-compatible endpoint, and the Tavily, LinkUp, Brave and Parallel search providers are paid separately, some with a free tier. SearXNG is the primary source and can be self-hosted from the `companions/` directory. The `local-inference` branch currently mirrors main and serves as a placeholder for swapping defaults. The client must be configured by hand in the MCP config and restarted. Installation is from source with `pip install -e .` and there is no registry package. Synthesis answers should be checked against the citations.
How to disable. Remove the `gigaxity-deep-research` block from `mcpServers` in `~/.claude.json` and restart the client. The clone folder and virtual environment can be deleted.
FAQ
How do I connect Gigaxity Deep Research to Claude Code?
Clone the repository, run `pip install -e .` in a virtual environment, fill `.env` and add a stdio block with `run_mcp.py` to `mcpServers` in `~/.claude.json`. After a restart `/mcp` shows the server.
Which tools does the server give?
Seven: `search`, `vertical_search`, `research`, `ask`, `discover`, `synthesize` and `reason`. In the agent they appear as `mcp__gigaxity-deep-research__search` and so on.
Which model is used for synthesis?
Qwen3-30B-A3B-Thinking through OpenRouter is recommended, but any OpenAI-compatible endpoint works: DeepSeek-R1, Qwen QwQ, Llama or a local vLLM, SGLang or llama.cpp.
Are paid keys required?
An OpenRouter key is needed for synthesis unless you run the model yourself. Search providers are paid separately, and SearXNG can be self-hosted from `companions/searxng/`.
Can it run as a REST API?
Yes, the same pipeline runs on FastAPI, and the `X-OpenRouter-Api-Key` header overrides the key per request so several users can pay from their own accounts.
Discussion
No questions or reviews yet.
Related
MCP serversGPT Researcher MCP
GPT Researcher MCP Server
An MCP server for deep web research: GPT Researcher vets sources on its own and hands the agent a ready report
Linkup MCP server: real-time web search, page fetching and long research tasks with sources for an agent
Metasearch for agents: one MCP server and HTTP API merge results from many search providers into unified JSON with deduplication.
MCP server for web search and crawling: SearXNG and You.com plus clean Markdown from pages, sites and logged-in areas
A CLI and MCP server that sends one question to search and AI sources and saves answers, citations and cost in a run.
A local MCP server with no API keys: multi-engine web search, page fetching and document reading tuned for LLMs.
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Authentication | Repository stars | Main difference |
|---|---|---|---|---|
| Gigaxity Deep Research (this card) | no | API key | 66 | |
| GPT Researcher MCP | yes | API key | 378 | a shim around GPT Researcher, whereas this has its own RRF result fusion and cited synthesis pipeline |
| Deep Research | no | 1,216 | a skill with a research procedure, whereas this is an MCP server with seven tools and a choice of model |
README in short
The README describes the multi-source search and synthesis pipeline, the seven tools, a quick start for MCP stdio and REST API, installation stages, branch options and architecture. Agent instructions live in `CLAUDE.md` and `AGENTS.md`.