uniprot-link
An MCP server that grounds protein research in the UniProt SPARQL endpoint, providing tools for querying proteins, sequences, variants, diseases, and more via intent-named tools and raw SPARQL.
README
uniprot-link
An MCP (Model Context Protocol) + REST server that grounds protein research in the
UniProt SPARQL endpoint (https://sparql.uniprot.org/sparql). It wraps a
~232-billion-triple, QLever-backed SPARQL 1.1 service behind intent-named,
token-economical tools — and ships a safe raw-SPARQL escape hatch plus UniProt's
126 curated example queries so an LLM can learn and write its own queries.
Part of the *-link family of biomedical MCP servers (gnomad-link, gtex-link,
pubtator-link, genereviews-link, …) and follows their stack and conventions.
Features
- 14 MCP tools across discovery, raw SPARQL, the curated example catalog, proteins (UniProtKB), and taxonomy.
- Typed protein tools —
get_protein, sequence, features (with FALDO coordinates), natural variants, diseases, GO terms, cross-references, id mapping. run_sparql_query— execute any SPARQL 1.1 query (SELECT/ASK/CONSTRUCT/ DESCRIBE, withSERVICEfederation); auto-LIMIT on unbounded SELECTs; JSON/XML/ CSV/TSV/Turtle/RDF-XML/N-Triples output.- Example catalog —
search_example_queries/get_example_queryexpose UniProt's 126 curated, executable queries (backed by thesparql-examplesgraph). - Agentic affordances — every response carries
_meta.next_commands(ready-to-run{tool, arguments}steps), a structured error taxonomy, and auniprot://capabilitiesdiscovery resource. - Three transports — unified (REST + MCP/HTTP), HTTP-only, and stdio.
Quick start
make install # uv sync --group dev
make dev # unified server: REST on / and MCP on /mcp (port 8000)
make mcp-serve # stdio MCP server (for Claude Desktop)
make ci-local # format + lint + loc + typecheck + tests
Connect Claude Desktop
Add to claude_desktop_config.json:
{
"mcpServers": {
"uniprot-link": {
"command": "uv",
"args": ["--project", "/path/to/uniprot-link", "run", "python", "mcp_server.py"]
}
}
}
Tool catalog
| Tool | Purpose |
|---|---|
get_server_capabilities |
Tools, 21 named graphs, prefixes, formats, workflows, limits |
run_sparql_query |
Execute any SPARQL 1.1 query (the power tool / federation) |
search_example_queries |
Search 126 curated example queries |
get_example_query |
Full text + metadata of one example |
find_proteins |
Search UniProtKB by gene / organism / keyword / EC / mnemonic |
get_protein |
Core entry summary by accession |
get_protein_sequence |
Canonical + isoform sequences |
get_protein_features |
Sequence features with FALDO coordinates |
get_protein_variants |
Natural-variant annotations |
get_protein_diseases |
Disease annotations |
get_protein_cross_references |
Cross-references grouped by database |
get_protein_go_terms |
GO annotations by aspect |
map_identifiers |
Map an accession to external DB ids |
get_taxon |
Resolve a taxon by id or name |
Configuration
Environment variables (prefix UNIPROT_LINK_, nested with __):
| Variable | Default | Meaning |
|---|---|---|
UNIPROT_LINK_SPARQL__CONTACT_EMAIL |
bernt.popp@charite.de |
Contact in the User-Agent (UniProt etiquette) |
UNIPROT_LINK_SPARQL__TIMEOUT |
30 |
Per-request timeout (s) |
UNIPROT_LINK_SPARQL__DEFAULT_LIMIT |
50 |
Auto-LIMIT for unbounded SELECTs |
UNIPROT_LINK_TRANSPORT |
unified |
unified / http / stdio |
UNIPROT_LINK_PORT |
8000 |
Server port |
Development
See AGENTS.md for conventions (notably the SPARQL / QLever discipline that
keeps queries off the timeout cliff) and docs/ for architecture and usage.
research/verify_queries.py validates every query builder against the live
endpoint.
Disclaimer
Research use only; not for clinical decision support, diagnosis, treatment, or patient management. UniProt data is licensed CC BY 4.0.
License
MIT
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。