🌱 SaltNet · Plant Salinity-Response Gene Database

All of SaltNet's underlying data is openly available in the GitHub repository. Use the links below to download individual files, or get the whole dataset at once.

SaltNet ships a local Model Context Protocol server so an AI agent (Claude Desktop, Claude Code, Cline, …) can query the database directly — fully offline, no backend or API key. Three tools:

1 · Install

git clone https://github.com/YunchuanWang/saltnet.git
cd saltnet/mcp_server
python -m venv .venv && source .venv/bin/activate   # Windows: .venv\Scripts\Activate.ps1
pip install -r requirements.txt

2 · Register with your agent

Use absolute paths, and point the command at the venv Python.

Claude Code:

claude mcp add saltnet -- /ABS/PATH/saltnet/mcp_server/.venv/bin/python /ABS/PATH/saltnet/mcp_server/saltnet_mcp.py

Claude Desktop — add to claude_desktop_config.json:

{
  "mcpServers": {
    "saltnet": {
      "command": "/ABS/PATH/saltnet/mcp_server/.venv/bin/python",
      "args": ["/ABS/PATH/saltnet/mcp_server/saltnet_mcp.py"]
    }
  }
}

3 · Ask

• “Use SaltNet to find Na⁺ transporters involved in salinity response in wheat.”
• “What does SaltNet say about SOS1 in Arabidopsis — evidence and which papers?”
• “Show SOS1's regulatory network and the experiment behind each edge.”

SaltNet was built from the literature by a multi-agent LLM pipeline. That pipeline is packaged as a second MCP toolbox — the build server — so you can construct your own evidence-traceable knowledge graph + database for any topic, with your own keywords and your own choice of models. Eight stages take you from a keyword to a finished resource: search → filter → extract → arbitrate → audit → consolidate → build → check.

Pipeline flowchart — stage → program → requirement

What each stage needs

Install

conda env create -f environment.yml && conda activate saltnet   # Python + EDirect in one go
#   or:  pip install -r requirements.txt        # then install EDirect separately
# Claude Code CLI (only for the LLM stages):  https://claude.com/claude-code

Run it — command line (your keywords)

make search QUERY='("cold stress"[TIAB]) AND plants[MeSH]' MINDATE=2000
make filter
make extract      # LLM · dual-model
make arbitrate    # LLM
make audit        # LLM
make consolidate  # merge → master data
make build        # db + KG + tables + figures
make check

Or drive the same stages conversationally via the MCP server below — heavy LLM stages run as background jobs (get_job / list_jobs) with a cost guardrail.

Register the build server (MCP, stdio)

{
  "mcpServers": {
    "saltnet-pipeline": {
      "command": "/home/<you>/miniconda3/envs/salt_nlp/bin/python",
      "args": [".../pipeline/mcp_pipeline/pipeline_mcp.py"]
    }
  }
}

Point command at the Python from the conda env above and args at your clone's pipeline/mcp_pipeline/pipeline_mcp.py (absolute paths).

Missing a gene, a correction, or an idea? Send it below. Submissions are reviewed before appearing on the public board; your email (optional) stays private.