SuperAPI AI Directory
Partner with us中文AI newsFavorites
All toolsAI developer platformsuniprot-mcp

uniprot-mcp

Query the UniProt knowledgebase (entries, features, variants, PTMs, GO terms, cross-references) with per-query SHA-256 provenance and offline replay.

Open website

About this tool

Purpose

uniprot-mcp is an auditable UniProt MCP server designed to generate verifiable, release-aware protein evidence bundles from UniProt and associated scientific sources. It helps users look up proteins, assemble protein, target, and variant evidence, and keeps an inspectable record of the source of every result.

Main capabilities

  • Source tracking and verification: Every successful response records the UniProt release version, retrieval time, resolved source URL, and SHA-256 digest.uniprot_provenance_verify Tools can later determine whether the upstream record is unchanged or has drifted.
  • Complete tool catalog: Provides eight endpoint families, all read-only (readOnlyHint: true), and except for uniprot_replay_from_cache , all interact with at least one upstream service (openWorldHint: true). No UniProt API key is required.
    • Core UniProtKB (10 tools)
    • Controlled vocabulary (4 tools)
    • Sequence archives and clustering (4 tools)
    • Proteomes and literature (4 tools)
    • Structured cross-database resolvers (4 tools, gateway only, does not leave the UniProt source)
    • Biomedical features (7 tools, pure Python composition, aimed at drug discovery, therapeutic protein engineering, pathogen secretion analysis)
    • Cross-source enrichment (3 tools, the only ones that consult sources outside rest.uniprot.org)
    • Composition and provenance (5 tools)
  • Release version pinning: Optional release version pinning is available for strict reproducibility.
  • Offline replay:uniprot_replay_from_cache(url) Reads previously recorded responses from the directory pointed to by UNIPROT_MCP_CACHE_DIR . Note: the cache must be populated by an external process; automatic write-through is not yet connected to the request path.
  • Pre-registered benchmark:tests/benchmark/ Provides an evaluation of 30 prompts (Tier A/B/C x 10), with expected answers committed via SHA-256, ensuring the authors cannot modify the correct answers after the fact.
  • Testing and quality: 1005 offline tests + 45 live integration tests, all passing. Line and branch coverage is 100.00% across all seven source files. mypy (strict), ruff, bandit, and pip-audit all report no issues.

How to use

  • Run with one command (recommended, no installation required):uvx uniprot-mcp-server
  • Install into an environment:pip install uniprot-mcp-server
    • Note: there is an unrelated uniprot-mcp package on PyPI (different author, 5 tools, MIT). Running pip install uniprot-mcp will silently install the wrong package. This package is uniprot-mcp-server.
  • Configure the MCP client: configure in claude_desktop_config.json . For pinned, reproducible-level access, release version pinning can be enabled.
  • Claude Code (CLI): supported.
  • Self-test: provides a live UniProt smoke check.

Example workflows

  1. Clinical variant interpretation package for TP53 R175H.
  2. Generate a drug target profile in a single call.
  3. Sequence chemistry for buffer selection/expression system selection.
  4. Source round-trip - prove the answer is reproducible.
  5. Offline replay of previously cached answers (read primitive).

Suitable scenarios

  • Biomedical researchers who need verifiable, reproducible protein evidence packages.
  • Clinical variant interpretation, such as TP53 R175H.
  • Drug target profile assembly.
  • Sequence chemistry analysis for buffer selection or expression system selection.
  • Regulated users who need audit trails and can prove prior answers are reproducible or detect upstream drift.

Limitations

  • All tools are read-only and do not provide the ability to write to or modify UnniProt.
  • uniprot_replay_from_cache is a read primitive; the cache must be populated by an external process (for example, a maintainer-provided baseline capture script, or the user wrapping httpx calls themselves and writing to the directory according to the documented JSON shape). Automatic cache write-through is not yet wired into the request path.
  • Cross-source enrichment tools consult sources outside rest.uniprot.org; see PRIVACY.md and the threat model.
  • The mutation testing infrastructure has been released, but the ≥95% threshold is a v1.2.0 target, not the current state.
  • The companion topologica-bio repository is currently private, under BUSL-1.1 (change date 2030-04-19, automatically falling back to Apache-2.0).

Citation and License

  • Cite via CITATION.cff (GitHub renders the "Cite this repository" button). Also cite the UniProt Consortium: The UniProt Consortium. UniProt: the Universal Protein Knowledgebase in 2025. Nucleic Acids Research (2025). doi:10.1093/nar/gkae1010
  • License: Apache-2.0, see LICENSE and NOTICE.
  • Copyright © 2026 Santiago Maniches. TOPOLOGICA LLC.

Related MCP Servers (same author)

  • alphafold-sovereign-mcp — integrates AlphaFold DB with eight additional public biomedical data sources, with a local SQLite knowledge graph (pip install alphafold-sovereign-mcp).
  • semantic-scholar-mcp — for Semantic Scholar (200 million+ academic papers), providing 14 tools for paper search, citation graph traversal, author profiles, and recommendations (pip install s2-mcp-server）。

Related tools

Browse category