ADMETlab MCP Server

A self-hostable MCP server that fronts the ADMETlab 3.0 web API so Claude can wash structures, render them, and predict ADMET (Absorption, Distribution, Metabolism, Excretion, Toxicity) properties in a drug-discovery workflow.

   
Type MCP server
Supplier ToxMCP (community)
Availability Beta — install-from-source, part of the ToxMCP suite
Pricing Free / OSS (Apache-2.0); backed by the public ADMETlab 3.0 API
Capabilities Read-only — submits SMILES to ADMETlab and returns predictions/renderings
Verified degraded · 2026-07-27 — repo resolves; uvicorn launch documented, not smoke-tested
Security caution · 2026-07-27 — provenance matches ToxMCP, Apache-2.0, no OSV advisories, early Beta 2-star

How to install

This is a long-lived HTTP MCP server — it does not run over stdio. You must start it in its own terminal and leave it running; Claude connects to it over HTTP. It is not on PyPI, so install from a clone.

  • Prerequisite — Python 3.10+ and Claude Code / Claude Desktop.

  • Clone, install, and start the server (keep this terminal open): 
    git clone https://github.com/ToxMCP/admetlab-mcp
cd admetlab-mcp
python -m venv .venv
source .venv/bin/activate
pip install -e ".[dev]"
cp .env.example .env
uvicorn admetlab_mcp.transport.http:app --host 127.0.0.1 --port 8200

    The MCP endpoint is then http://127.0.0.1:8200/mcp (health check at http://127.0.0.1:8200/health). Leave this process running while you use the tool.

  • Claude Code — register the running HTTP endpoint (in a second terminal): 
    claude mcp add --transport http admetlab http://127.0.0.1:8200/mcp
  • Claude Desktop — Claude Desktop has no native HTTP transport, so proxy the HTTP endpoint through mcp-remote in claude_desktop_config.json: 
    {
  "mcpServers": {
    "admetlab": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "http://127.0.0.1:8200/mcp"]
    }
  }
}

    (Requires Node.js for npx; the uvicorn server above must already be running.)

What it does

Exposes four tools that front the ADMETlab 3.0 API:

  • wash_molecule — standardize/clean an input molecule (POST /api/washmol).
  • render_molecule_svg — render a structure to SVG (POST /api/molsvg).
  • predict_admet — run ADMET property prediction for a molecule (POST /api/single/admet, with a documented fallback to /api/admet).
  • fetch_admet_csv — retrieve prediction results as CSV by task ID (POST /api/admetCSV).

Batching is supported (up to ~1000 SMILES per prediction call) with rate limiting (≤5 rps), retries/backoff, and fallback endpoints.

Primary use cases: ADMET property prediction, compound triage during lead optimization, structure washing and rendering for drug-discovery pipelines.

Notes

No API key is required (an optional placeholder for future authentication is included). Predictions depend on the upstream ADMETlab 3.0 service, which the project notes can be unstable — expect occasional 5xx/404 responses passed through from upstream. Because this is an HTTP/SSE server, it must be kept running in a separate terminal; this is the most common cause of “the tool doesn’t show up” for HTTP MCP servers.

For a zero-install, enterprise-hosted ADMET path in Claude.ai, see the Inductive Bio ADMET Connector; for broader cheminformatics property modeling see PyTDC and DeepChem.

Sources

Installed this tool?

Share feedback — install path, OS, errors, workarounds. The form opens with this tool pre-selected and a link back to this page.