ENCODE Toolkit is genomics research infrastructure for the ENCODE Project, usable as an MCP server and a Claude Code plugin. Its purpose is to let researchers complete the entire workflow from data retrieval to analysis method generation in natural language within Claude.
Purpose It is mainly used to search, download, track, and analyze functional genomics experimental data in the ENCODE Project. According to the project introduction, using traditional genomics databases requires browsing web portals, clicking through numerous filters, manually searching for experiments and files across multiple databases, writing custom scripts for batch downloads, and is prone to losing file provenance information. ENCODE Toolkit allows users to describe their needs directly to Claude, for example "find all histone ChIP-seq data for human pancreatic tissue," and Claude will search ENCODE and return a structured table, with downloads organized by experiment, accompanied by MD5 checksums and complete provenance tracking.
Main capabilities
- Data retrieval and browsing: supports searching ENCODE experiments, with more than 20 filters available.
- Data download and tracking: supports downloading specified files by accession, searching and downloading, tracking experiments and their publications, methods, and pipeline information, listing experiment files, searching files across experiments, and retrieving file metadata.
- Database cross-referencing: the project states it integrates 14 databases and can cross-reference through real-time API tools and guided skills, for example cross-comparing histone peaks with GWAS variants, checking ClinVar pathogenicity, obtaining GTEx expression, and analyzing transcription factor binding motifs from JASPAR.
- Analysis and pipelines: provides 7 Nextflow pipelines covering analyses such as ChIP-seq and ATAC-seq, and includes ENCODE standard QC thresholds.
- Method and provenance generation: can record derived files, view the provenance chain from derived files back to source ENCODE data, and generate methodology sections with citations.
- Export and citation: can export tracked experiments as CSV, TSV, or JSON; retrieve publications and export them as BibTeX or RIS.
- External reference linking: can associate external references such as PubMed, bioRxiv, ClinicalTrials, and GEO.
Tools and skills The project provides 20 tools, including encode_search_experiments, encode_get_experiment, encode_download_files, encode_batch_download, encode_track_experiment, encode_list_files, encode_search_files, encode_get_metadata, encode_get_facets, encode_get_file_info, encode_manage_credentials, encode_list_tracked, encode_get_citations, encode_compare_experiments, encode_summarize_collection, encode_log_derived_file, encode_get_provenance, encode_export_data, encode_link_reference, encode_get_references. When installed as a Claude Code plugin, it also includes 47 literature-supported workflow skills. Each analysis skill includes evidence-based quality thresholds, assay-specific metrics, and primary literature citations. Each pipeline includes a SKILL.md overview, 5-stage reference files covering preprocessing through QC, a complete Nextflow DSL2 pipeline, a Dockerfile, and deployment configurations for local, SLURM, GCP, and AWS.
Usage Recommended installation is as a Claude Code plugin: enter the command in a new Claude Code session. If you only need the 20 MCP tools and not the 47 workflow skills, you can also configure it separately. In the MCP client configuration, you can use { "command": "npx", "args": ["encode-toolkit"] }. In Claude Desktop, add the configuration to claude_desktop_config.json: on macOS the path is ~/Library/Application Support/Claude/claude_desktop_config.json, and on Windows the path is %APDATA%\Claude\claude_desktop_config.json. When using uvx, no installation is needed; just add the configuration and restart Claude. It can also be added to the workspace's .vscode/mcp.json, .cursor/mcp.json, or .windsurf/mcp.json. Most ENCODE data is public and requires no authentication. For restricted or unpublished data, you can have Claude store ENCODE credentials. Credentials are encrypted using the operating system keychain (macOS Keychain, Linux Secret Service, Windows Credential Locker) and are not stored in plaintext. Supported file formats include: fastq, bam, bed, bigWig, bigBed, tsv, csv, hic, tagAlign, bedpe, pairs, fasta, vcf, tar.
Use cases
- Find ChIP-seq, ATAC-seq, RNA-seq, and other experiments for a specific tissue, cell line, or species.
- Batch download and organize files such as BED, FASTQ, bigWig, and perform MD5 verification.
- Track experiments and their publications, and export citations.
- Cross-reference ENCODE data with databases such as GWAS, ClinVar, GTEx, JASPAR, and others.
- Set up and run Nextflow pipelines for analyses such as ChIP-seq and ATAC-seq.
- Record the provenance of derived files and generate methodology sections.
Limitations and notes The project states 100% local execution, with no telemetry, analytics, or tracking; credentials are encrypted via the operating system keychain, with Fernet as a fallback; certificate verification is enforced; rate limiting is in place to comply with ENCODE's policy of 10 requests per second; MD5 verification is performed by default on all downloads; data does not leave the local machine except through HTTPS queries to public APIs. Before version 0.3.1, the mcp dependency was not constrained, causing new installations to pull the incompatible mcp 2.x; this issue was fixed in 0.3.1. If you encounter issues, you can run uvx --refresh encode-toolkit or uv cache clean encode-toolkit to refresh the cache; pip users can run pip install --upgrade encode-toolkit. The license is AGPL-3.0. For commercial licensing inquiries, contact ammawla@ucdavis.edu. Citation information: Alex M. Mawla. (2026). ENCODE-Toolkit: an MCP server, Claude plugin, and skills suite for ENCODE genomic data access and analysis. Zenodo. https://doi.org/10.5281/zenodo.18917511
