Synthesize Bio

Generate synthetic human gene expression data from a plain-language experiment description and run differential…

Add to Agentman Studio

Opens Agent Studio, where connecting is one click. The connector URL below works in any MCP client.

Paste it into any MCP client. Setup docs

Resources

What the Synthesize Bio connector does

The Synthesize Bio MCP server connects a generative gene expression model to Claude, ChatGPT and any MCP-compatible agent. Describe an experiment in plain language — "tumor vs normal lung tissue" — and the platform generates synthetic human expression profiles and runs differential expression on them. Nothing it returns is a measurement. Sign-in is OAuth with one analyze scope.

Verified connector

Listed by Anthropic as a partner connector in its Connectors Directory.

Connection checked by Agentman on .

Anthropic states this reflects the level of review a connector received, not a security audit.

Synthesize Bio tools (6)

  • resolve_sample_metadata
  • analyze_gene_expression
  • get_analysis_results
  • get_saved_analysis_report
  • annotate_genes
  • get_counts_data_url

Tool names from Anthropic's directory listing. This server requires sign-in, so we could not read tool descriptions or parameter schemas.

Limits

  • The data is generated, not measured. GEM-1 predicts expression from metadata. Every figure, fold change and p-value downstream of this connector describes model output, and any use that does not say so is a misrepresentation waiting to happen.
  • One pairwise comparison per run, and extra groups vanish silently. Synthesize Bio documents that with three or more groups the analysis compares the first two alphabetically and ignores the rest. It is a note in the docs, not an error at runtime.
  • The vendor documents four tools; Anthropic lists six. get_saved_analysis_report and annotate_genes have no published description, no parameter schema and no stated effect.
  • No published price, and a throttle the vendor may change without notice. No pricing or credits page exists; the terms permit limiting API calls and file sizes at the company's sole discretion, without notice, with suspension as the remedy.
  • Analyses are slow by design. Three to five minutes is the vendor's own typical figure, with server-side polling waits of up to about forty seconds each. This is not a connector to put behind an interactive prompt that expects an immediate answer.
  • We could not read tool schemas or safety annotations. The endpoint returned 401 to an anonymous request, so no readOnlyHint or destructiveHint is published here, and the two undocumented tools have no parameter detail from any source.
  • The analyze scope draws no internal boundary. One scope covers resolving metadata, starting a compute job and minting download URLs alike. There is nothing to grant narrowly.
  • Presigned download URLs are bearer credentials. The counts URL is valid for one hour to anyone holding it. Do not paste it into a shared channel.
  • Metadata curation is not fully reviewed. Synthesize Bio states that it does not manually review every harmonized metadata result and advises checking metadata when you need high confidence in a specific sample or study.
  • Quality flag cutoffs on the underlying public data are deliberately liberal. Synthesize Bio flags samples below 50% aligned reads, above 80% duplication, or under 10,000 non-zero genes, and says explicitly that you should compare those against expectations for your protocol.
  • Single-cell support is narrower than bulk. The sampling_strategy option Synthesize Bio documents for realistic measurement error is marked bulk-only in its SDK reference.
  • No medical or clinical disclaimer appears in the terms. We looked. The absence is worth knowing given the marketing positions the product around clinical trial design.

Frequently asked questions

Does the Synthesize Bio connector read my own patient or genomic data?

No. Nothing in the tool surface uploads or reads a file from you. You describe an experiment in words, and the platform generates gene expression profiles from its own model. Synthesize Bio documents that its bulk training data comes from the NCBI Sequence Read Archive, a public repository, so the reads behind the model are already public rather than yours.

Is the gene expression data the Synthesize Bio connector returns real?

No, it is model output. Synthesize Bio describes GEM-1 as a generative model that predicts expression profiles from metadata alone, and its Python SDK calls the result synthetic expression data. The numbers look like a sequencing run and are not one. Treat every result as a hypothesis to test at the bench, not as a measurement.

How many tools does the Synthesize Bio MCP server actually have?

Anthropic's directory lists six and Synthesize Bio's own reference documents four. The vendor page states plainly that the server exposes four tools and covers resolve_sample_metadata, analyze_gene_expression, get_analysis_results and get_counts_data_url. Two directory names, get_saved_analysis_report and annotate_genes, appear nowhere in the vendor documentation.

Does running a Synthesize Bio analysis cost money or credits?

Synthesize Bio does not publish a price. Its documentation carries no pricing, credit or quota page, and its terms of service describe subscriptions, automatic renewal and fees without naming what the MCP tools consume. Its terms also let the company cap API calls at its sole discretion and without notice. Check your account before looping an agent over analyses.

Why does a Synthesize Bio analysis take so long in Claude?

Because it runs a real inference pipeline in the background. Synthesize Bio documents that most runs take about three to five minutes and that the flow is asynchronous by design: the analysis returns a job ID immediately and Claude polls it. Each poll waits server-side for up to roughly forty seconds. Slowness here is expected behavior, not a failed request.

Can the Synthesize Bio connector compare more than two groups at once?

No. Synthesize Bio's documentation states that the analysis performs one pairwise comparison per run, and that if a prompt implies three or more groups the current behavior compares the first two alphabetically while the rest are ignored. That silent discard is the failure mode to watch. Run additional comparisons as separate prompts.

What OAuth permission does the Synthesize Bio MCP server request?

A single scope named analyze. The server's RFC 9728 resource descriptor and its authorization server metadata both advertise exactly that one scope, checked 2026-08-22. It is a real application scope rather than an identity claim, but with only one value there is nothing to grant separately, so consent cannot separate resolving metadata from running the pipeline.

Can the Synthesize Bio connector delete anything or write to my account?

No listed name deletes. The six directory names cover resolving, analyzing, retrieving and annotating, and Anthropic's directory entry records the connector's permissions as read only. The counterpoint is that analyze_gene_expression starts a compute job and creates a platform dataset, so the connector does add state to your account even without a delete verb.

Sources

More Healthcare connectors

  • AdisInsight

    Query Springer Nature's drug-pipeline and trial database from an AI agent.

  • AdvancedMD

    Look up patients, find open slots, book appointments, and work denials and remittances in AdvancedMD from your AI agent.

  • AllTrails

    Search AllTrails' trails by place, map bounds or name and pull details and a 7-day trailhead forecast from your AI…

Browse all MCP connectorsClaude and ChatGPT setup guideBrowse AI skills

Product names and logos belong to their owners. Agentman is not affiliated with a connector's maker unless the page says the connector is built and run by Agentman.

Use it in an agent

Put Synthesize Bio to work.

Connect Synthesize Bio once and your agents call these tools on their own: on a schedule, in a workflow, with nobody at the keyboard.

Open in Agentman Studio