Synthesize Bio MCP Server

Generative genomics MCP server. Predict and compare gene expression between sample groups from natural language prompts, powered by the GEM-1 foundation model.

Healthcare & Life Sciences by Synthesize Bio OAuth2 active

Connect (Remote Server) Docs

Overview

Synthesize Bio provides a hosted MCP server that exposes their generative genomics platform to AI agents. The backend is powered by GEM-1, a foundation model trained on a deeply curated RNA-seq corpus that generates in silico gene expression data from experimental design descriptions. Through the connector, an AI client like Claude can resolve a natural language description of an experiment into structured sample groups, run a differential expression analysis, and return statistically tested results, all without the user touching the platform UI.

The server exposes five tools that map to the platform's asynchronous job flow: resolving sample metadata, kicking off an analysis, polling for results, downloading raw counts, and annotating Ensembl gene IDs. Analyses cover bulk and single-cell RNA-seq modalities and apply Welch's t-test with Benjamini-Hochberg correction, returning up to 1,000 significant genes per job. Jobs typically finish in 3 to 5 minutes.

The MCP is available as an official Claude Connector via Claude's marketplace using OAuth 2.0 with PKCE, and also works with any MCP-compatible client (such as Cursor) using a Bearer token API key. It is best suited for biologists, computational genomics researchers, and AI-assisted drug discovery workflows that need on-demand transcriptomic comparisons.

Tools

Tool	Description
`resolve_sample_metadata`	Interpret a natural-language experiment description into structured sample groups for bulk or single-cell RNA-seq.
`analyze_gene_expression`	Start the differential expression analysis pipeline using a confirmed metadata resolution. Returns a job_id for polling.
`get_analysis_results`	Poll job status and return progress or final results, including up to 1,000 significant genes with Welch's t-test and Benjamini-Hochberg correction.
`get_counts_data_url`	Return a presigned download URL (valid for 1 hour) for the raw expression counts data produced by a completed job.
`annotate_genes`	Map Ensembl gene IDs to gene symbols and synonyms. Limit of 500 IDs per call.

Setup Guide

Synthesize Bio hosts the MCP server, so there is nothing to install or run locally. Authenticate either through Claude's connector marketplace (OAuth) or with an API key for custom clients.

Prerequisites

A Synthesize Bio account at app.synthesize.bio
For API key auth: create a key at app.synthesize.bio/account/api-keys

Option A: Claude Connector (recommended)

Open Claude Settings, then Connectors
Search the marketplace for "Synthesize Bio"
Complete the OAuth sign-in flow with your Synthesize Bio account
Confirm the connector appears as connected

Option B: Custom MCP client (Cursor, etc.)

Server URL:

https://app.synthesize.bio/api/mcp

Auth header:

Authorization: Bearer YOUR_API_KEY

Example client config:

{
  "mcpServers": {
    "synthesize-bio": {
      "url": "https://app.synthesize.bio/api/mcp",
      "headers": {
        "Authorization": "Bearer YOUR_API_KEY"
      }
    }
  }
}

Use Cases

Run quick differential expression comparisons (e.g. heart vs liver, tumor vs adjacent normal) directly from chat without touching a bioinformatics pipeline.
Generate synthetic control samples or rare condition data for experimental design and power analysis.
Triage candidate genes for follow-up wet lab work by asking the agent to summarize the top up- and down-regulated genes.
Annotate Ensembl IDs returned by other tools into human-readable gene symbols inside an agent workflow.
Pair with literature search MCPs to ask "what does this gene do?" after pulling significant hits from an analysis.

Example Prompts

"Compare gene expression between heart and liver cells and show the top up-regulated genes."
"Analyze lung adenocarcinoma tumor vs normal tissue in bulk RNA-seq."
"Compare CD4+ T cells vs CD8+ T cells in single-cell mode."
"Summarize the top up- and down-regulated genes from my last analysis in plain English."
"Annotate these Ensembl IDs: ENSG00000141510, ENSG00000157764."

Pros

Official, provider-hosted server with first-party Claude Connector and OAuth flow.
Backed by GEM-1, a domain-specific generative genomics foundation model, not a generic LLM guessing biology.
Real statistical rigor: differential expression uses Welch's t-test with Benjamini-Hochberg correction.
Covers both bulk and single-cell RNA-seq modalities through the same conversational interface.

Limitations

Narrow scope: limited to differential expression between two sample groups, not full multi-condition or trajectory analyses.
Results are in silico generated, not wet lab measurements, so they require validation for downstream decisions.
Requires a Synthesize Bio account and is subject to platform quotas; asynchronous jobs take 3 to 5 minutes per request.

Alternatives

BioMCP and the bio-mcp collection of bioinformatics MCP servers for tool-based workflows (BLAST, sequence analysis, PubMed).
pysynthbio or rsynthbio: Synthesize Bio's own Python and R API wrappers, for non-MCP programmatic access to the same platform.
Custom MCP wrappers around public expression atlases like GTEx, Expression Atlas, or the Human Cell Atlas for real (non-generated) data.