LabscoMCP early access ↗
Augmented-NatureMCP SERVER

AlphaFold MCP Server

Pull AlphaFold predicted structures by UniProt accession — PDB, CIF, BCIF or JSON — with per-residue confidence analysis, batch retrieval, and ready-to-run PyMOL and ChimeraX scripts.

by Augmented-Nature ↗Biomedical: Genomics & Clinical DataCommunity

LABSCO SUMMARY

Confidence is treated as data, not decoration.

The per-residue scores, the region breakdown and the quality verdict are separate tools from the structure fetch, which is the right split — most mistakes with AlphaFold models come from trusting a low-confidence loop as if it were solved. The exports carry that colouring straight into PyMOL or ChimeraX, so what you see in the viewer matches what the analysis said.

What it is

An MCP server over the AlphaFold Protein Structure Database API at https://alphafold.ebi.ac.uk/api/. It retrieves and downloads predicted structures by UniProt accession, reports per-residue confidence and where the low-confidence regions are, processes proteins in batches, compares several structures at once, and emits visualisation scripts. It caches nothing — every call goes to the official API.

What you get

  • A structure by UniProt accession in pdb, cif, bcif or json, with an availability check before you ask — get_structure, download_structure, check_availability
  • Structures found by protein or gene name, listed for an organism, and organism-level coverage statistics — search_structures, list_by_organism, get_organism_stats
  • Per-residue confidence scores with an optional threshold, a breakdown of high- and low-confidence regions, and the prediction's version, date and quality metrics — get_confidence_scores, analyze_confidence_regions, get_prediction_metadata
  • Batch work with explicit caps: structure info for up to 50 accessions, downloads for up to 20, confidence analysis for up to 30 — batch_structure_info, batch_download, batch_confidence_analysis
  • Side-by-side comparison of 2 to 10 structures, plus structurally related proteins for a reference accession — compare_structures, find_similar_structures
  • Sequence coverage and an overall quality verdict for a prediction — get_coverage_info, validate_structure_quality
  • Visualisation handoff: PyMOL and ChimeraX exports with confidence colouring on by default, and an API status check — export_for_pymol, export_for_chimerax, get_api_status
  • Resource templates for direct addressing — alphafold://structure/{uniprotId}, alphafold://pdb/{uniprotId}, alphafold://confidence/{uniprotId}, alphafold://summary/{organism}

Requirements

Nothing — no account, no key; AlphaFold's public API needs neither. Build from a clone with npm install and npm run build, then point the client at build/index.js with node. The package is alphafold-server 1.0.0 and speaks stdio. Batch operations are capped per call and large requests are chunked automatically with built-in delays, which is the server's way of staying inside the API's tolerance.

Setup effort

Build from source — git clone https://github.com/Augmented-Nature/AlphaFold-MCP-Server

See installation options