UniProt MCP Server

An MCP (Model Context Protocol) server that provides tools for querying the UniProt protein database using the unipressed Python library.

Features

• Search proteins using standard the UniProt query syntax across UniProtKB, UniParc, or UniRef databases
• Fetch specific entries by accession ID
• Pagination for large result sets
• Field selection to control returned data
• JSON format responses by default - responses are returned in JSON format, with TOON format available as an option

Installation

Using uv (recommended)

uv sync

This will create a virtual environment and install all dependencies.

Development installation

uv sync --extra dev

This installs the package with development dependencies (pytest, etc.).

Usage

Running the server

With uvx (recommended)

uvx uniprot-unipressed-mcp

Or from a git repository:

uvx --from git+https://github.com/pansapiens/uniprot-unipressed-mcp uniprot-mcp

Alternative example: with the FastMCP CLI and HTTP transport

fastmcp run src/uniprot_mcp/server.py:mcp --transport http --port 8007

Configuring with MCP clients

Claude Code

Add the server using the Claude Code CLI:

claude mcp add uniprot-unipressed-mcp -- uvx git+https://github.com/pansapiens/uniprot-unipressed-mcp

Other MCP clients

Or manually add to your MCP configuration (Claude Code, Cursor, etc.):

{
  "mcpServers": {
    "uniprot-unipressed-mcp": {
      "command": "uvx",
      "args": ["git+https://github.com/pansapiens/uniprot-unipressed-mcp"]
    }
  }
}

Response Format

Tool responses are returned in JSON format by default. To receive TOON format responses instead, use the response_format parameter with value "toon":

# JSON format (default)
result = uniprot_search(query="gene:BRCA1")

# TOON format
result = uniprot_search(query="gene:BRCA1", response_format="toon")

Tools

uniprot_search

Search the UniProt protein database using query syntax.

Parameters:

Parameter	Type	Required	Default	Description
query	string	Yes	-	UniProt query string
database	string	No	"uniprotkb"	Database to search: uniprotkb, uniparc, uniref
limit	integer	No	10	Results per page (1-100)
fields	list[string]	No	None	Return fields to include
cursor	string	No	None	Pagination cursor from previous result
response_format	string	No	"json"	Response format: "json" (default) or "toon"

Example queries:

gene:BRCA1                              # Search by gene name
organism_id:9606                        # Human proteins (NCBI taxonomy ID)
(gene:BRCA*) AND (organism_id:10090)    # Mouse BRCA genes with wildcard
length:[500 TO 700]                     # Proteins of specific length range
keyword:kinase                          # By UniProt keyword
family:serpin                           # By protein family
ec:3.2.1.23                             # By enzyme classification
reviewed:true                           # Only Swiss-Prot reviewed entries

Response:

By default, responses are returned in JSON format. When response_format="toon" is specified, responses are returned in TOON format (a compact string):

JSON format (default):

{
  "results": [...],
  "total": 1234,
  "nextCursor": "eyJvZmZzZXQiOiAxMH0="
}

TOON format:

results[10]:
  - entryType: UniProtKB reviewed (Swiss-Prot)
    primaryAccession: P38398
    secondaryAccessions[7]: E9PFZ0,O15129,Q1RMC1,Q3LRJ0,Q3LRJ6,Q6IN79,Q7KYU9
    uniProtkbId: BRCA1_HUMAN
    ...etc...

uniprot_fetch

Fetch specific protein entries by their UniProt accession IDs.

Parameters:

Parameter	Type	Required	Default	Description
ids	list[string]	Yes	-	UniProt accession IDs to fetch
database	string	No	"uniprotkb"	Database to fetch from
fields	list[string]	No	None	Return fields to include
response_format	string	No	"json"	Response format: "json" (default) or "toon"

Example:

uniprot_fetch(ids=["P62988", "A0A0C5B5G6"])

Response:

By default, responses are returned in JSON format. When response_format="toon" is specified, responses are returned in TOON format (a compact string):

JSON format (default):

{
  "results": [...],
  "found": 2,
  "requested": 2
}

TOON format:

results[10]:
  - entryType: UniProtKB reviewed (Swiss-Prot)
    primaryAccession: P38398
    secondaryAccessions[7]: E9PFZ0,O15129,Q1RMC1,Q3LRJ0,Q3LRJ6,Q6IN79,Q7KYU9
    uniProtkbId: BRCA1_HUMAN
    ...etc...

Pagination

The server uses cursor-based pagination for search results. When more results are available, the response includes a nextCursor field. Pass this cursor in subsequent requests to retrieve the next page:

# First request
result1 = uniprot_search(query="organism_id:9606", limit=10)

# Get next page using cursor
if "nextCursor" in result1:
    result2 = uniprot_search(
        query="organism_id:9606",
        limit=10,
        cursor=result1["nextCursor"]
    )

Databases

Database	Description
uniprotkb	UniProt Knowledgebase - curated protein sequences and annotations
uniparc	UniProt Archive - comprehensive protein sequence archive
uniref	UniProt Reference Clusters - clustered protein sequences

Return Fields

Common return fields include:

• accession - UniProt accession number
• id - Entry name
• gene_names - Gene names
• protein_name - Protein names
• organism_name - Source organism
• organism_id - NCBI taxonomy ID
• length - Sequence length
• mass - Molecular mass
• sequence - Amino acid sequence
• cc_function - Function annotation
• cc_subcellular_location - Subcellular location

See the UniProt return fields documentation for the complete list.

Development

Using a local copy of the repository

{
  "mcpServers": {
    "uniprot-unipressed-mcp": {
      "command": "uv",
      "args": ["--directory", "/path/to/uniprot-unipressed-mcp", "run", "-m", "uniprot_mcp.server"]
    }
  }
}

Testing using the MCP Inspector

npx @modelcontextprotocol/inspector uv --directory $(pwd) run -m uniprot_mcp.server

Running tests

uv run pytest

This runs all unit tests. Integration tests (which require network access to the UniProt API) are skipped by default.

Running integration tests

Integration tests can be enabled using an environment variable:

RUN_INTEGRATION_TESTS=1 uv run pytest

Or run only integration tests:

RUN_INTEGRATION_TESTS=1 uv run pytest -m integration

Running tests with coverage

uv run pytest --cov=uniprot_mcp

To include integration tests in coverage:

RUN_INTEGRATION_TESTS=1 uv run pytest --cov=uniprot_mcp

Resources

• UniProt
• UniProt Query Syntax
• UniProt Return Fields
• unipressed Documentation
• FastMCP Documentation
• Model Context Protocol
• TOON Format - Alternative serialization format available as an option

Licence

MIT