Skip to main content
Glama
Sign In
rnaseq-mcp

rnaseq-mcp

by sbarbi-gh
OverviewSchemaSourceRelated ServersDiscussionsAdmin
Python
Local

rnaseq-mcp

Server MCP per l'analisi di dati Bulk RNA-seq tramite agenti LLM. Permette di interrogare un dataset in linguaggio naturale. Esegue codice R e Python in un container Docker isolato, con anonimizzazione automatica dei campioni e controllo sulla privacy degli output.

Prerequisiti

  • Docker

  • Python ≥ 3.11

  • Un client MCP compatibile (es. Claude Code, Mistral Vibe o qualsiasi agente che supporti il protocollo MCP)

Related MCP server: DROMA MCP Server

Quickstart

1. Build del container

cd container
docker build -t rnaseq-executor:latest .

Nel container vengono installati pacchetti di R (DESeq2, edgeR, limma, clusterProfiler, fgsea) e le librerie di annotazione (org.Hs.eg.db, org.Mm.eg.db) per garantire funzionalita offline.

2. Installazione del server

pip install -e .

3. Configurazione del client MCP

Registra il server nel tuo client MCP con questi parametri:

Parametro

Valore

Comando

rnaseq-mcp

COUNTS_FILE

percorso assoluto a counts.csv

METADATA_FILE

percorso assoluto a metadata.csv

I file di input consistono in 2 CSV:

  • counts.csv — matrice geni × campioni (gene ID come indice di riga)

  • metadata.csv — tabella campioni × variabili (stesso indice di counts.csv)

4. Utilizzo

Una volta connesso il server, i tool disponibili sono:

Tool

Descrizione

list_data_files

Elenca i file in /data

read_table

Anteprima di un CSV

execute_r

Esegue codice R (sessione persistente)

execute_python

Esegue codice Python (namespace persistente)

get_outputs

Recupera i risultati da /output

describe_session

Stato della sessione e pacchetti disponibili

reset_session

Azzera container, stato e output

Esempio di sessione:

Analizza i dati RNA-seq in /data confrontando treated vs control,
correggi per batch e genera volcano plot e heatmap.

Viene fornito un esempio prompt di sistema in prompts/system_prompt.md.

Privacy

All'avvio, il server:

  1. Rimuove le colonne di metadati con identificatori personali (nomi, ID campione, barcode, ecc.)

  2. Rinomina i campioni con alias anonimi S1, S2, ..., SN

  3. Blocca l'esportazione di matrici per-campione (count, PCA) come CSV

Mappatura dei campioni

La corrispondenza S1 → nome_reale viene scritta su host nel file:

/tmp/rnaseq-<session_id>/host_only/sample_mapping.json

Il file non è montato nel container e non è accessibile tramite i tool MCP — il modello non può mai risalire ai nomi originali. Il formato è:

{ "S1": "nome_originale_1", "S2": "nome_originale_2", ... }

Il file viene eliminato alla chiusura del server (o alla chiamata di reset_session). Per conservarlo, copialo manualmente prima di terminare la sessione:

cp /tmp/rnaseq-*/host_only/sample_mapping.json ./sample_mapping.json

Output

I file scritti in /output dal codice R/Python sono montati su host nella directory:

/tmp/rnaseq-<session_id>/output/

Sono accessibili direttamente dal filesystem dell'host durante tutta la sessione, senza dover passare per il tool get_outputs. Quest'ultimo è utile principalmente per restituire i risultati al modello (ad es. per leggere una tabella CSV o visualizzare un plot nel contesto della conversazione).

I file vengono eliminati alla chiusura del server o alla chiamata di reset_session. Per conservarli:

cp -r /tmp/rnaseq-*/output/ ./risultati/

Pacchetti R disponibili

DESeq2, edgeR, limma, ggplot2, pheatmap, ComplexHeatmap, EnhancedVolcano, clusterProfiler, fgsea, msigdbr, org.Hs.eg.db, org.Mm.eg.db, AnnotationDbi

Licenza

MIT License — Copyright (c) 2026 Stefano Barbi

Available Tools

7 tools
describe_sessionA

Return R sessionInfo(), available Python symbols, and files already produced in /output. Useful at the start of a session to confirm what packages are available.

Inputs: no named parameters. Returns: result

execute_pythonA

Execute Python code in the isolated analysis container. The Python namespace is PERSISTENT across calls.

OUTPUT RULES (same as execute_r):

  • Write to /output ONLY aggregated results (gene/pathway level).

  • Do NOT write per-sample matrices as CSV.

  • Plots saved with matplotlib/seaborn: plt.savefig("/output/.png").

  • If the server blocks a CSV, produce an aggregated equivalent instead.

Available packages: pydeseq2, pandas, numpy, matplotlib, seaborn, scipy, statsmodels.

Inputs: code (required). Returns: result

execute_rA

Execute R code in the isolated analysis container. The R session is PERSISTENT: objects defined in earlier calls remain available.

OUTPUT RULES (enforced server-side — follow them in your code):

  • Write to /output ONLY aggregated results: DE tables (log2FC, pvalue, padj per gene), pathway scores, group-level summary statistics, model coefficients.

  • Do NOT write per-sample matrices (normalized counts, raw counts, PCA coordinates) as CSV/TSV to /output. Keep them in R memory for intermediate computation.

  • Plots (PNG) showing sample-level data are permitted — use anonymous labels (S1, S2, ...).

  • ggsave() and png() write to /output/; use descriptive names.

  • If the server blocks a CSV file, revise your code to export an aggregated version.

Available R packages: DESeq2, edgeR, limma, ggplot2, pheatmap, ComplexHeatmap, EnhancedVolcano, ggrepel, patchwork, clusterProfiler, fgsea, msigdbr, org.Hs.eg.db, org.Mm.eg.db, AnnotationDbi. No internet access — use org.Hs.eg.db instead of biomaRt for gene annotation.

Inputs: code (required). Returns: result

get_outputsA

Retrieve all files written to /output during this session. CSV files containing per-sample data are blocked and replaced with an explanatory message. Images (PNG) are returned as base64-encoded strings.

Inputs: no named parameters. Returns: result

list_data_filesA

List files available in /data (anonymized counts and metadata).

Inputs: no named parameters. Returns: result

read_tableA

Read the first n_rows of a CSV/TSV file from /data and return it as a markdown table. Useful for inspecting column names, sample aliases, and group labels before writing analysis code.

Inputs: filename (required), n_rows. Returns: result

reset_sessionA

Destroy the current Docker container and start a fresh one with the same anonymized data. Use this to clear all in-memory R/Python state and /output files.

Inputs: no named parameters. Returns: result

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    Deploy Server
    One endpoint for every MCP toolRayrun runs every MCP server behind one URL and holds the credentials. Each call is decided per client, tool, and argument, then recorded.ray.runAd

    TDQS

    A4.4/5.0

    Scored across 7 tools

    Disambiguation5/5

    Each tool has a clear, distinct purpose: environment inspection, code execution per language, output retrieval, data exploration (listing and reading), and session reset. No two tools overlap in functionality.

    Naming Consistency5/5

    All tool names follow a consistent verb_noun pattern (e.g., describe_session, execute_python, list_data_files). The naming is predictable and easy to interpret.

    Tool Count5/5

    With 7 tools, the server covers all essential activities for RNA-seq analysis: environment setup, code execution, data exploration, output retrieval, and session management. The count is well-scoped and not excessive.

    Completeness4/5

    The tool surface provides a solid foundation for RNA-seq analysis: data listing, reading, execution in both R and Python with relevant packages, output retrieval, and reset. A minor gap is the lack of a direct quality control or standard pipeline tool, but the execution tools suffice.

    Maintenance

    ActivityInactive
    ResponsivenessNo recent community issues

    Resources

    Related MCP Connectors

    Related MCP Servers

    • SCMCP

      huang-sh
      A
      license
      C
      quality
      D
      maintenance
      An MCP server that enables single-cell RNA sequencing analysis through natural language, supporting data processing, visualization, and analysis tasks without requiring coding knowledge.
      52
      12
      BSD 3-Clause

    • DROMA MCP Server

      DROMA01
      A
      license
      Not graded
      quality
      D
      maintenance
      Enables natural language interactions with cancer pharmacogenomics data through the DROMA platform, supporting drug-omics association analysis, dataset management, molecular profile loading, and treatment response analysis across multiple research projects.
      1
      Mozilla Public 2.0

    • SCMCP

      scmcphub
      A
      license
      Not graded
      quality
      D
      maintenance
      An MCP server that enables scRNA-Seq analysis through natural language, providing tools for data preprocessing, clustering, and biological visualization. It supports both predefined function execution and a flexible code mode powered by a Jupyter backend for automated single-cell transcriptomics workflows.
      16 PyPI
      16
      BSD 3-Clause

    • decoupler-MCP

      scmcphub
      F
      license
      Not graded
      quality
      D
      maintenance
      Provides a natural language interface for single-cell RNA-Seq analysis using the decoupleR framework. It enables users to perform biological pathway inference, data clustering, and visualization through MCP-compatible AI clients.
      31 PyPI
      4
      -