Download skills/bionemo-genmol-nim/SKILL.md from the publisher, or copy the path and open the repository. PlugMCP lists the skill and does not host the file.
Where does genmol-nim live?
The publisher keeps it at https://github.com/NVIDIA/skills.
Which agents can use genmol-nim?
Claude, Cursor, and Codex can read a SKILL.md you add from the publisher's repository.
>
name: genmol-nim
description: >
Generate novel drug-like molecules using the GenMol NIM microservice. Use for de novo generation, scaffold decoration, motif extension, lead optimization, SAFE notation, QED or LogP ranking, hosted NVIDIA API calls, or local Docker deployment. GenMol takes SAFE notation in the smiles field, not ordinary SMILES.
license: Apache-2.0 AND CC-BY-4.0
compatibility: "safe-mol>=0.1.14; requests>=2.28"
allowed-tools: Bash, Read, Write, AskUserQuestion
GenMol NIM
Generate drug-like molecules with GenMol. Use this guide for first-pass
hosted and local usage; load supplemental files only when needed:
Hosted requests use Authorization: Bearer $NGC_API_KEY. For local Docker,
authenticate image pulls with docker login nvcr.io using NGC_API_KEY
(or NVIDIA_API_KEY via the preflight). Pass -e NGC_API_KEY into the
container for entitlement checks and first-run model downloads. Local inference
requests use no auth header after readiness, so bind the published port to
loopback with -p 127.0.0.1:8000:8000. Warm-cache key-free startup varies by
image version and should not be assumed.
Local Docker
Use credentials already supplied in the shell environment or injected by a
secret manager. Do not load credential files, print keys, or enable shell tracing.
For local setup answers, include this sequence: env preflight, docker login
with --password-stdin, docker run, readiness loop, then a no-auth localhost
request. Do not invent a cache default or drop the NVIDIA_API_KEY fallback.
Before executing local setup, explain that registry authentication sends the key
to the NVIDIA registry at https://nvcr.io and first-run model downloads use
about 20 GB in LOCAL_NIM_CACHE.
Execute deployment only when the user requests it; for a setup guide, provide
the commands without running them.
For the exact startup preflight (environment checks, NVIDIA_API_KEY fallback,
--shm-size=2G, both --ulimit flags, docker login, and the docker run
for nvcr.io/nim/nvidia/genmol:1.0.1), copy the command block in
references/api.md under Local container startup verbatim.
GenMol is single-GPU; NIM_TEST_GPU defaults to 0. Wait for readiness:
until curl -sf http://localhost:8000/v1/health/ready; do sleep 5; done
SAFE Input
The API field is named smiles, but GenMol expects SAFE notation. Masked
positions use [*{min-max}].
Wider masks increase diversity; tight masks keep analog size more predictable.
Request Pattern
import os
import requests
HOSTED = True
url = (
"https://health.api.nvidia.com/v1/biology/nvidia/genmol/generate"
if HOSTED else "http://localhost:8000/generate"
)
headers = {"Content-Type": "application/json"}
if HOSTED:
headers["Authorization"] = f"Bearer {os.getenv('NGC_API_KEY')}"
payload = {
"smiles": "[*{20-30}]", # SAFE notation
"num_molecules": 30,
"temperature": "1", # string, not float
"noise": "1", # string, not float
"step_size": 1,
"scoring": "QED", # or "LogP"
"unique": False,
}
response = requests.post(url, headers=headers, json=payload, timeout=180)
response.raise_for_status()
result = response.json()
Gotchas:
temperature and noise are strings.
num_molecules is 1-1000; invalid/duplicate molecules may be filtered, so
request extra when the user needs a minimum count.
scoring is "QED" for drug-likeness or "LogP" for lipophilicity.
Set unique=True for deduplicated analog lists.
Save And Report Output
Sort molecules by score, print the top ranks, and write a .smi file as shown
in references/examples.md under Save Ranked
Results. For chemical validity, uniqueness, PAINS/alerts, and visualization
with RDKit, read references/validation.md.
Limits And Troubleshooting
Fewer molecules than requested is expected after filtering.
Invalid SAFE strings cause status: "failed" or validation errors.
Install safe-mol only for scaffold, motif, or lead-optimization workflows;
de novo masks work without conversion.
Local startup downloads about 20 GB into LOCAL_NIM_CACHE.
Container issues: confirm nvidia-smi, NVIDIA Container Toolkit, and
--runtime=nvidia; use NIM_TEST_GPU to choose the single visible GPU.
How do I install genmol-nim?
Download skills/bionemo-genmol-nim/SKILL.md from the publisher, or copy the path and open the repository. PlugMCP lists the skill and does not host the file.
Where does genmol-nim live?
The publisher keeps it at https://github.com/NVIDIA/skills.
Which agents can use genmol-nim?
Claude, Cursor, and Codex can read a SKILL.md you add from the publisher's repository.