使用说明

Litmus MCP Server

Model Context Protocol (MCP) server for Litmus experiment intake, validation, routing, and submission.

Overview

This MCP server allows AI assistants (Claude, ChatGPT, etc.) to:

Draft experiment intakes from natural language
Validate intakes against the schema
Route experiments to best-fit labs
Submit experiments and manage attachments

Installation

pip install litmus-mcp

Or install from source:

git clone https://github.com/litmus-science/litmus-mcp.git
cd litmus-mcp
pip install -e .

Running the Server

litmus-mcp

Or run as a module:

python -m litmus_mcp.server

Configuration

Claude Desktop

Add to your Claude Desktop config (~/Library/Application Support/Claude/claude_desktop_config.json on macOS):

{
  "mcpServers": {
    "litmus": {
      "command": "litmus-mcp"
    }
  }
}

Claude Code

Add to your Claude Code MCP settings:

{
  "mcpServers": {
    "litmus": {
      "command": "litmus-mcp"
    }
  }
}

Tools

intake.draft_from_text

Convert natural language to a structured intake JSON.

{
  "text": "I want to test if my new compound kills cancer cells",
  "experiment_type_hint": "CELL_VIABILITY_IC50"
}

intake.validate

Validate an intake against the schema.

{
  "intake": { ... },
  "strict": false
}

intake.suggest_questions

Get questions to improve intake completeness.

{
  "intake": { ... },
  "target_completeness": 0.7,
  "max_questions": 3
}

routing.match_labs

Route an intake to best-fit labs.

{
  "intake": { ... },
  "top_k": 3,
  "strict_deliverables": false,
  "region_preference": "US"
}

labs.search

Search labs with filters.

{
  "experiment_type": "CELL_VIABILITY_IC50",
  "max_bsl": "BSL2",
  "region": "US"
}

labs.get

Get a specific lab profile.

{
  "lab_id": "lab_cellassay_pro_001"
}

intake.submit

Submit a validated intake.

{
  "intake": { ... },
  "selected_lab_id": "lab_cellassay_pro_001"
}

files.create_upload_url

Create a signed URL for file upload.

{
  "filename": "compound_sds.pdf",
  "mime_type": "application/pdf",
  "size_bytes": 102400
}

files.attach_to_intake

Attach an uploaded file to a submission.

{
  "submission_id": "sub_abc123",
  "file_id": "file_xyz789",
  "attachment_type": "SDS"
}

Resources

The server exposes these resources:

litmus://schemas/experiment_intake - Experiment intake JSON Schema
litmus://schemas/lab_profile - Lab profile JSON Schema
litmus://taxonomy/deliverables - Deliverables taxonomy
litmus://rubric/routing_weights - Current routing weights

Development

# Clone the repo
git clone https://github.com/litmus-science/litmus-mcp.git
cd litmus-mcp

# Install dev dependencies
pip install -e ".[dev]"

# Run tests
pytest

# Test with MCP Inspector
npx @modelcontextprotocol/inspector litmus-mcp

License

MIT