The ordering constraint is written into the descriptions — search first for Ensembl IDs, then info, seq, bgee and opentargets take them — which is the difference between an agent that carries the right identifier forward and one that guesses at a symbol. The width has a price: some tools answer only after gget_setup has installed their dependencies, and the heaviest query in the set comes with a hardware footnote rather than a spinner.
An MCP surface over the gget bioinformatics toolkit, 20 tools spanning gene search, sequence retrieval, alignment, structure prediction and public expression databases.
- Gene symbols resolved to Ensembl IDs in batches, which the metadata, sequence and association tools then require
- Gene and transcript metadata for those IDs, and sequences written to a local file, optionally translated
- Reference genome and annotation FTP links by species
- Similarity search and alignment: BLAST against a BLAST database, BLAT against a UCSC assembly, MUSCLE and DIAMOND alignments saved to file
- Expression and association data from ARCHS4, Bgee, OpenTargets and CZ CELLxGENE Census, plus functional enrichment through Enrichr
- Structures fetched from PDB or predicted with AlphaFold, and linear motifs predicted locally from a sequence or UniProt ID
- COSMIC cancer-mutation search, and sequences mutated according to standard annotation
No key. Third-party dependencies for the modules you intend to use, which gget_setup installs per module, and a writable output_path for the tools that save sequences, alignments and structures. gget_cellxgene's own note puts large Census queries above 16 GB of RAM and a 5 Mbps connection.
One command — uvx gget-mcp stdio