MCP server for the GigaChat API
GigaChat API
A small Python MCP server for Sber's GigaChat API: authorizes with a Basic key and calls the models directly from an agent, tested on OpenCode under Windows
Medium risk
We rate an entry medium when the tool runs code, makes network calls or reads project files. Check what exactly it does before installing.
Why this level
- It spends paid GigaChat tokens under the user's key
- The Basic base64 auth key is stored and transmitted by the server
Install
Manual install
pip install -r requirements.txtInstall Python dependencies.
This is third-party code. Review the repository files before installing.
What it does
The server sets up an MCP wrapper over the GigaChat API: on first run it asks for an auth key in Basic base64 format and then works as a regular MCP server for an agent. The repository also ships a Snippets folder with example code from Sber itself (generated via Postman), which the server does not actually use and keeps only as reference material for the API's capabilities: async batch generation, embeddings, file handling, image recognition, AI-text detection, and token balance monitoring.
Who it is for. For people who want to quickly connect GigaChat to an agent over MCP with no cloud intermediary.
Good fit when
- You need direct GigaChat API access from an agent like OpenCode
- You have a GigaChat Basic auth key and want a minimal server with no extra scaffolding
Not a fit when
- You need a mature, tested implementation with tool documentation: the server's README is very short
- You have no access key for the paid GigaChat API
Example request
Generate a GigaChat response to this request and show the remaining token balanceLimitations
The top-level README is very short; the tool list and parameters are not documented there, with details deferred to mcp_server/README.md. The Snippets folder with Sber's examples is not used by the server itself, it is just reference material the author left in. It was developed and tested for running on Windows.
How to disable. Stop the server process (start_mcp.bat or python server.py) and remove it from your MCP client config.
MCP
- Transport
- stdio
- Authentication
- API key
| Environment variables | |
|---|---|
| GIGACHAT_AUTH_KEY required, secret | GigaChat auth key in Basic base64 format, requested on the server's first run |
Security check
- It spends paid GigaChat tokens under the user's key
- The Basic base64 auth key is stored and transmitted by the server
README in short
The README gives the repository layout, a three-step quick start (install dependencies, run the server, enter the auth key on first run), and a pointer to a separate README inside mcp_server for details, noting the Snippets folder is only Sber's sample code and is not used by the server.
FAQ
Does the server use the Snippets folder?
No, those are just Sber's sample code the author kept as a development reference.
Does it work outside Windows?
The README also gives a Linux/Mac command (python server.py), but the main example is a Windows batch file.
Related
Open-source personal AI assistant on your own machine: answers in Telegram, Slack, Discord and WhatsApp, extended with skills and plugins
A self-improving agent from Nous Research with a TUI, messaging gateway, cron jobs and skills it writes itself
An open source coding agent for the terminal and desktop with build and plan modes
Open prompt library with a Claude Code plugin, MCP server and CLI: search, fetch and improve prompts and skills from an agent