Encode Toolkit

What ammawla/encode-toolkit MCP server does

ammawla/encode-toolkit MCP server connects an MCP-compatible client to the ENCODE Project for genomic data discovery and local research organization. Its tools cover experiment searches, experiment details, file listings, file metadata, downloads, and filter metadata. The server can also maintain a local tracker for experiments and record relationships between source ENCODE files and derived outputs.

The repository describes support for common functional genomics data types, including ChIP-seq, ATAC-seq, RNA-seq, Hi-C, WGBS, and CUT&RUN. Beyond ENCODE’s live API tools, the project includes skills that guide work with databases such as GTEx, ClinVar, GWAS Catalog, JASPAR, CellxGene, gnomAD, Ensembl, UCSC Genome Browser, GEO, PubMed, bioRxiv, ClinicalTrials.gov, and Open Targets. These connected systems are presented as skills or MCP integrations rather than as the core ENCODE tool set.

How it works

Search tools accept experiment and file filters, while metadata and facet tools help identify valid filter values and show available-data counts. After an experiment is selected, the client can retrieve its files, quality information, audit details, publications, methods, and pipeline information. Downloads can target selected accessions or combine a search with a download operation; batch download starts in preview mode by default.

Local tracking supports collection-level workflows. Tracked experiments can be compared for combined-analysis compatibility, summarized statistically, exported as CSV, TSV, or JSON, and used to produce BibTeX or RIS citations. External references from sources including PubMed, bioRxiv, ClinicalTrials.gov, and GEO can be linked to tracked experiments. Derived files can be logged with their source data so provenance chains can be inspected later.

Setup and configuration

The README documents a Python 3.10-or-newer project and provides a direct MCP-only command using uvx:

uvx encode-toolkit

Claude Desktop, Cursor, and Windsurf configurations use the same uvx command with encode-toolkit as its argument. The README also documents installation through pip and an npx-based configuration. A Claude Code plugin is available through the project’s marketplace commands and includes the MCP connector plus workflow skills.

The server includes a credential-management tool for storing, checking, or clearing ENCODE credentials when restricted data access is needed. The supplied material does not specify environment variables or a required credential format, so configuration details for authenticated access should be checked in the project documentation.

Tools and capabilities

The documented MCP tools support:

• Searching experiments and files with combined filters.
• Retrieving experiment, file, audit, quality, and facet metadata.
• Listing files by experiment, format, or type.
• Downloading individual files or previewing batch downloads.
• Tracking experiments and listing tracked collections.
• Exporting tracked data as CSV, TSV, or JSON.
• Comparing experiments for combined analysis.
• Exporting tracked publications as BibTeX or RIS.
• Logging derived files and inspecting provenance chains.
• Linking and retrieving external references.

Limitations and notes

The project is marked beta in the supplied README. Its live MCP tools are centered on ENCODE, while many additional databases are accessed through guided skills or separate MCP integrations. The material does not establish that every listed database is directly exposed as an MCP tool by this server. Restricted ENCODE data may require credentials, but public-data access and the exact authentication requirements are not specified here.

ammawla/encode-toolkit MCP server is licensed under AGPL-3.0. The README also identifies the project as having no telemetry and documents a full provenance workflow, but users should verify current behavior and version-specific details before relying on those properties in a production research process.