Official STRING Database MCP Server
# Official STRING Database MCP Server
[](https://mcpqueen.com/s/org.string-db/string-mcp) [](https://glama.ai/mcp/servers/meringlab/string-mcp)
Exposes [STRING database](https://string-db.org) functionality as a **Model Context Protocol (MCP)** server.
This implementation allows AI agents and other MCP-compatible clients to access STRING data through a structured and self-describing interface.
It is built on top of the STRING API but adapted specifically for model-based use, with emphasis on conciseness and context efficiency. The server controls the amount and type of data so that responses stay within the reasoning limits of large language models. It also adapts the structure and adds metadata to support more consistent parsing and interpretation by agentic systems.
The server is also available as a hosted remote MCP endpoint at [https://mcp.string-db.org](https://mcp.string-db.org).
For a live demonstration of MCP capabilities, see [STRING Chat](https://string-db.org/cgi/chat).
For reproducible workflows and large-scale integration, use the [STRING API](https://string-db.org/cgi/help?subpage=api).
For agentic workflows, use the optional agent skill at [skills/string-database/SKILL.md](skills/string-database/SKILL.md).
## Features
- Resolve protein identifiers to STRING metadata
- Retrieve interaction networks
- Cluster and visualize interaction networks
- Perform homology lookups across species
- Access evidence links for protein–protein interactions
- Run functional enrichment analysis (including enrichment plots)
- Get curated functional annotations for proteins
- Query proteins by function or disease (GO, KEGG, Reactome, DISEASES etc.)
---
## Requirements
- **Python** ≥ 3.11 (with Python 3.10, optional parameter descriptions are misplaced in the tool schemas)
- Dependencies (see `requirements.txt`):
- `fastmcp==2.14.7`
- `mcp==1.29.0`
- `httpx>=0.28,<0.29`
- `pydantic>=2.11,<3.0`
> **Note:** If the server crashes on startup, it is very likely due to an incompatible **FastMCP** version.
---
## Configuration
The server reads settings from `config/server.config` (JSON). Example:
```json
{
"base_url": "https://string-db.org",
"server_port": 57416,
"verbosity": "low"
}
```
## Installation
```bash
git clone https://github.com/meringlab/string-mcp.git
cd string-mcp
pip install -r requirements.txt
```
## Running
```
python server.py
```
To serve over standard input/output instead of HTTP, for example when a local MCP client starts the server itself:
```
python server.py --stdio
```
The stdio mode also reads `config/server.config` for the STRING API endpoint.
## Testing
```bash
curl -v -H "Accept: text/event-stream" http://localhost:[port]/mcp
```
You should see a `200 OK` response with `content-type: text/event-stream`.
**Note:** HTTP requests sent with default headers will trigger `anyio.ClosedResourceError` and `406 Not Acceptable`. This is expected behavior for the MCP streaming protocol and does not indicate a problem.
## Docker
1. Ensure you have a `config/server.config` file (copy from `config/server.config.example` if needed):
```bash
cp config/server.config.example config/server.config
```
2. Build the image:
```bash
docker build -t string-mcp .
```
3. Run the container:
**Linux/macOS:**
```bash
docker run -p 57416:57416 -v $(pwd)/config:/app/config:ro string-mcp
```
**Windows PowerShell:**
```powershell
docker run -p 57416:57416 -v ${PWD}/config:/app/config:ro string-mcp
```
**Windows Command Prompt:**
```cmd
docker run -p 57416:57416 -v %cd%/config:/app/config:ro string-mcp
```
To run in detached mode:
```bash
# Linux/macOS
docker run -d -p 57416:57416 -v $(pwd)/config:/app/config:ro --name string-mcp-server string-mcp
# Windows PowerShell
docker run -d -p 57416:57416 -v ${PWD}/config:/app/config:ro --name string-mcp-server string-mcp
```
The `-v` flag mounts your local config directory so you can customize settings without rebuilding the image.
**Note:** If you don't need to modify the config, you can omit the volume mount and the container will use the default config from the image.
4. Stop the container (if running in detached mode):
```bash
docker stop string-mcp-server
docker rm string-mcp-server
```
**Note:** The Docker container will use the configuration from `config/server.config`. If this file doesn't exist, the container will copy from `server.config.example` on first run.
Frozen STRING releases are available under versioned paths: `/12.0/` (STRING v12.0) and `/12.5/` (STRING v12.5).
## License / Citation
The STRING MCP server source code is released under the [MIT License](https://opensource.org/licenses/MIT).
Associated data and outputs are released under the [CC BY 4.0 License](https://creativecommons.org/licenses/by/4.0/).
You are free to use, share, and modify the code with proper attribution.
If you use this code or data in your work, please also cite the latest [STRING manuscript](https://string-db.org/cgi/about?footer_active_subpage=references).
TDQS
Scored across 17 tools
Most tools target distinct actions (resolve, query-set interactions, all partners, clustering, homology, enrichment, annotation, term/sequence/species search). The main overlap is between string_visual_network and string_network_link, which both describe nearly identical single/multi-protein network construction, though their descriptions clarify image vs interactive URL output.
All tools share a predictable string_ prefix and snake_case, which is highly consistent. However the ordering convention varies (verb_noun like resolve_proteins/sequence_search/create_file vs noun_phrases like interactions_query_set, network_link, proteins_for_term), a minor deviation.
17 tools is on the heavier side but justified by the genuinely broad STRING surface (resolution, interactions, networks, clustering, enrichment, annotations, homology, files, help). Each tool maps to a distinct STRING capability rather than being redundant.
The surface comprehensively covers the STRING domain: identifier resolution, interaction queries, partner discovery, network/link/clustering generation, homology, evidence pages, functional enrichment, annotation, term and sequence search, species lookup, file export, and help. No obvious dead ends for typical STRING workflows.