LLM Bridge MCP
A unified interface to various LLM providers like OpenAI, Anthropic, and Google using Pydantic AI.
LLM Bridge MCP
LLM Bridge MCP allows AI agents to interact with multiple large language models through a standardized interface. It leverages the Message Control Protocol (MCP) to provide seamless access to different LLM providers, making it easy to switch between models or use multiple models in the same application.
Features
- Unified interface to multiple LLM providers:
- OpenAI (GPT models)
- Anthropic (Claude models)
- Google (Gemini models)
- DeepSeek
- ...
- Built with Pydantic AI for type safety and validation
- Supports customizable parameters like temperature and max tokens
- Provides usage tracking and metrics
Tools
The server implements the following tool:
run_llm(
prompt: str,
model_name: KnownModelName = "openai:gpt-4o-mini",
temperature: float = 0.7,
max_tokens: int = 8192,
system_prompt: str = "",
) -> LLMResponse
prompt: The text prompt to send to the LLMmodel_name: Specific model to use (default: "openai:gpt-4o-mini")temperature: Controls randomness (0.0 to 1.0)max_tokens: Maximum number of tokens to generatesystem_prompt: Optional system prompt to guide the model's behavior
Installation
Installing via Smithery
To install llm-bridge-mcp for Claude Desktop automatically via Smithery:
npx -y @smithery/cli install @sjquant/llm-bridge-mcp --client claude
Manual Installation
- Clone the repository:
git clone https://github.com/yourusername/llm-bridge-mcp.git
cd llm-bridge-mcp
- Install uv (if not already installed):
# On macOS
brew install uv
# On Linux
curl -LsSf https://astral.sh/uv/install.sh | sh
# On Windows
powershell -ExecutionPolicy ByPass -c "irm https://astral.sh/uv/install.ps1 | iex"
Configuration
Create a .env file in the root directory with your API keys:
OPENAI_API_KEY=your_openai_api_key
ANTHROPIC_API_KEY=your_anthropic_api_key
GOOGLE_API_KEY=your_google_api_key
DEEPSEEK_API_KEY=your_deepseek_api_key
Usage
Using with Claude Desktop or Cursor
Add a server entry to your Claude Desktop configuration file or .cursor/mcp.json:
"mcpServers": {
"llm-bridge": {
"command": "uvx",
"args": [
"llm-bridge-mcp"
],
"env": {
"OPENAI_API_KEY": "your_openai_api_key",
"ANTHROPIC_API_KEY": "your_anthropic_api_key",
"GOOGLE_API_KEY": "your_google_api_key",
"DEEPSEEK_API_KEY": "your_deepseek_api_key"
}
}
}
Troubleshooting
Common Issues
1. "spawn uvx ENOENT" Error
This error occurs when the system cannot find the uvx executable in your PATH. To resolve this:
Solution: Use the full path to uvx
Find the full path to your uvx executable:
# On macOS/Linux
which uvx
# On Windows
where.exe uvx
Then update your MCP server configuration to use the full path:
"mcpServers": {
"llm-bridge": {
"command": "/full/path/to/uvx", // Replace with your actual path
"args": [
"llm-bridge-mcp"
],
"env": {
// ... your environment variables
}
}
}
License
This project is licensed under the MIT License - see the LICENSE file for details.
Похожие серверы
Salesforce MCP Server
Integrates Claude with Salesforce, enabling natural language interactions with your Salesforce data and metadata.
Commvault
Interact with Commvault software to access job details, security posture, and SLA status of the commcell.
Azure AHDS FHIR MCP Server
An MCP server implementation for interacting with Azure Health Data Services FHIR.
Salesforce MCP
Interact with Salesforce orgs securely and modularly using LLM-driven tools.
DigitalOcean
Provides comprehensive access to all DigitalOcean API endpoints, dynamically extracted from their OpenAPI specification.
AWS Cost Explorer & Bedrock Logs
Retrieve AWS spend data from Cost Explorer and Amazon Bedrock usage data from CloudWatch logs.
Garmin Connect
Access Garmin Connect running data and training plan information.
Paddle Billing
Interact with the Paddle Billing API to manage products, prices, customers, transactions, and subscriptions.
Kaggle
Interact with the Kaggle API to access datasets, notebooks, and competitions.
HubSpot MCP Server by CData
A read-only MCP server for querying live HubSpot data, powered by CData.