Claude Code life sciences marketplace: what it actually installs

The repository is a catalogue, not the servers themselves

The first thing to understand about anthropics/life-sciences is what it is not. Its own description says it is a repo for the Claude Code Marketplace, that it will continue to host the marketplace.json file long term, but not the actual MCP servers. So the code you see here is packaging and pointers: a `.claude-plugin` directory, a `.github` directory, a README, and then one directory per plugin.

That structure is visible in the top level listing. Directories named `pubmed`, `biorender`, `synapse`, `wiley-scholar-gateway`, `consensus`, `cortellis`, `adisinsight`, `10x-genomics`, `single-cell-rna-qc`, `instrument-data-to-allotrope`, `nextflow-development`, `scvi-tools` and `scientific-problem-selection` sit alongside others that the README does not document, including `biorxiv`, `chembl`, `clinical-trial-protocol-skill`, `clinical-trials`, `medidata`, `open-targets`, `owkin` and `tooluniverse`. The gap between the documented set and the directory list is the first thing worth flagging, since a plugin you find in the tree is not necessarily one the documentation explains.

The two plugin kinds behave differently. Remote MCP servers talk to an external service over the network. Local MCP servers, described as MCPB, run on your machine. Skills are not servers at all; they are domain instructions that extend what Claude can do inside a session, and they need no credentials.

PubMed is the only server that works with no account at all

Authentication is the axis that decides how much of this marketplace you can actually use, and the documentation is explicit about the tiers.

Only PubMed is listed under no authentication, described as accessible to all users, giving search and access to biomedical literature and research articles. Next come the free account group: BioRender for scientific illustrations and diagrams, Synapse.org for collaborative research data management from Sage Bionetworks, Wiley Scholar Gateway for academic research and publications, and Consensus, which is presented as AI-powered search across more than 200 million peer-reviewed papers with evidence synthesis. Consensus still requires an account even though it is grouped as free.

The paid group is where most life sciences teams will hit a wall. Cortellis from Clarivate covers regulatory intelligence on submissions, approvals and guidance documents. AdisInsight from Springer Nature covers drug development pipelines, clinical trials, safety and deals. Both need subscriptions. The 10x Genomics Cloud plugin is also listed under paid or institutional access, but it carries an extra condition that matters more than the price: the documentation notes it is only useful if you already have analysis data in your account. Installing it gives you nothing without that data.

Adding the marketplace and installing plugins from Claude Code

Installation happens through Claude Code slash commands, not a package manager. The quick start adds the marketplace by its GitHub slug, then installs individual plugins by their plugin ID:

# Add the marketplace
/plugin marketplace add anthropics/life-sciences

# Install MCP servers
/plugin install pubmed@life-sciences
/plugin install biorender@life-sciences
/plugin install synapse@life-sciences
/plugin install wiley-scholar-gateway@life-sciences
/plugin install 10x-genomics@life-sciences

# Install skills
/plugin install single-cell-rna-qc@life-sciences
/plugin install instrument-data-to-allotrope@life-sciences
/plugin install nextflow-development@life-sciences
/plugin install scvi-tools@life-sciences
/plugin install scientific-problem-selection@life-sciences

What you should see after these commands is a set of installed plugins inside Claude Code. The detailed installation section offers a longer form that clones the marketplace by URL instead, which is useful if you want to pin a fork or inspect the manifest before adding it:

/plugin marketplace add https://github.com/anthropics/life-sciences.git

Install only what you need. Every plugin you add is a new integration that Claude can call, and each one you cannot authenticate is a dead entry in your plugin list. The naming convention is consistent, with the plugin ID carrying the marketplace suffix, for example `pubmed@life-sciences`.

Credentials are entered through the plugin menu, then Claude Code restarts

Configuration for authenticated servers is done inside Claude Code itself. The documented sequence is to type `/plugin`, select "Manage plugins", find the installed server, choose "Configure", enter the required credentials, and then restart Claude Code to activate the MCP servers. Where a server does not expose a configure option, the alternative path is to authenticate through the server's own web interface when prompted.

That last point is worth pausing on. "Configure (if available)" appears in the detailed steps, so there is at least one server where this menu route may not be present and the web flow is the only route. Restarting is not optional in either case, since the servers are not live until Claude Code starts again.

Nothing in the documentation explains where credentials are stored on disk, how they are encrypted, or whether a credential can be revoked centrally. If your institution requires that API keys for a commercial data product be held in a secret manager with rotation, you will have to make that arrangement yourself. For the 10x Genomics plugin the credential is an access token generated from the account security page of the vendor's cloud console, which means the token is minted in the vendor's UI and pasted into Claude Code.

The nextflow skill is the one that changes your workstation

Most plugins in this marketplace talk to somebody else's server. The nextflow-development skill runs work on your own machine, which makes its requirements the heaviest in the set: Docker and Nextflow installed locally.

The skill covers nf-core pipelines for `rnaseq` (gene expression and differential expression analysis), `sarek` (germline and somatic variant calling for WGS and WES data) and `atacseq` (chromatin accessibility analysis). The documented features are downloading public datasets from GEO or SRA, auto-detecting data types and suggesting appropriate pipelines, generating pipeline-compatible samplesheets, and giving environment validation and troubleshooting guidance. The target audience is stated plainly: bench scientists who need large scale omics analyses without specialised bioinformatics training.

That last phrase is the trade-off. The skill lowers the barrier, but it also means container orchestration and a workflow engine become part of your local setup, and nf-core pipelines pull their own containers and reference data. A machine that can run large genomics workloads is not a machine you hand to everyone. Two other skills sit in the same local-execution category in spirit: single-cell-rna-qc performs MAD-based filtering following scverse best practices with detailed visualisations, and scvi-tools covers deep learning analysis including model selection guidance, training workflows and integration pipelines for scVI, scANVI, totalVI, PeakVI and MultiVI.

Two skills are format conversion and research planning

Two entries in the catalogue are worth separating out because neither is a data connection at all.

The instrument-data-to-allotrope skill converts instrument data to the Allotrope Simple Model format, described as a standardised format for data exchange and analysis. It is a format bridge for standardized data exchange, and it needs no credentials. That makes it the cheapest thing in the marketplace to try, and the easiest way to see whether skills are useful in your workflow before paying for any subscription.

The scientific-problem-selection skill is a framework for choosing research problems and making strategic research decisions, based on the methodology of Fischbach and Walsh from Cell in 2024. Its listed use cases are pitching and refining new research ideas, evaluating project risks and feasibility, working through decision trees in active projects, and strategic planning. The honest framing is that this is a thinking aid with a citation behind it, not an instrument. It will not look at your data, and it carries no dependency on the other plugins, so it can be installed and ignored with no cost.

Licensing is delegated to each server, and the repository lists none

The license question is unusually ambiguous here. The license for the repository itself is recorded as unknown, and the README's own license section begins by saying that individual MCP servers come with their own terms but is cut off before finishing the sentence. So the wrapper code and the marketplace manifest in this repository have no clearly stated license, while each underlying service carries the licensing of its vendor.

That has a practical consequence. The BioRender, Cortellis, AdisInsight, Consensus and Synapse plugins grant access to commercial products under those vendors' terms, and a subscription is not a licence to redistribute what you pull out of it. The repository says nothing about that boundary, and it is not the right place to look for it. Check the terms of each service you connect, and check separately what licence, if any, covers the manifest and the plugin wrappers in this repository before you fork it into an internal marketplace.

On maintenance, the last push was on 2026-08-14, and the release history shows v1.1.1 on 2026-01-09 after v1.0.0 on 2025-10-20. The gap between those two releases is roughly three months, and the repository is not archived. Support boundaries are stated: problems with the Claude Code plugin system go to the `#claude-cli-feedback` channel on Anthropic Slack, while problems with a particular MCP server go to that provider.

Editorial conclusion

This marketplace is worth installing for a bench scientist who wants Claude Code to read PubMed and run nf-core pipelines without hand wiring MCP servers. It is the wrong choice for a team that needs a single governed data path, since credentials for Cortellis, AdisInsight and 10x Genomics live in individual subscriptions that the repository does not grant. Before adding anything, check the marketplace directory listing against your own accounts, confirm Docker and Nextflow are present if you plan to use the nextflow-development skill, and read the license section of the plugin you install since each MCP server carries its own terms and the repository license is not listed.

Frequently asked questions

Official sources

1. anthropics/life-sciences on GitHub
2. Issues
3. README
4. Releases