evalsmith-mcp
An open-source MCP server and CLI tool designed to evaluate RAG retrieval quality, agent tool call correctness, and prompt regressions using deterministic metrics. It supports integration with desktop MCP clients, IDEs, and CI/CD pipelines via GitHub Actions, providing structured reports and merge gates without requiring paid API keys.
Package and access
Local package with source-attributed installation metadata. Koskamo does not execute packages during review.
What you can do with it
- Evaluate RAG retrieval metrics like recall@k and nDCG
- Compare AI agent configurations with confidence intervals
- Check if agent tool calls match expected traces
- Run prompt regression tests against golden sets
Source and maintenance
- Maintenance checked
Packages and installation
Source-provided package data. Koskamo has not executed or live-probed this package.
evalsmith-mcp 0.1.0
{
"version": "0.1.0",
"transport": "stdio",
"identifier": "evalsmith-mcp",
"registryType": "npm",
"environmentVariables": [
{
"name": "EVALSMITH_JUDGE_BASE_URL",
"format": "string",
"default": "http://localhost:11434/v1",
"description": "OpenAI-compatible endpoint for the optional LLM judge (default: Ollama on localhost)"
},
{
"name": "EVALSMITH_JUDGE_MODEL",
"format": "string",
"default": "llama3.1:8b",
"description": "Model used by the optional LLM judge"
},
{
"name": "EVALSMITH_JUDGE_API_KEY",
"format": "string",
"description": "API key for the judge endpoint, if it needs one"
},
{
"name": "LOG_LEVEL",
"format": "string",
"default": "info",
"description": "pino log level; logs go to stderr"
}
]
}Tools
Reported by the attributed source; not independently observed by Koskamo.
Connect from your client
Each snippet configures the same source-provided package; replace placeholders before use.
Cursor
{
"mcpServers": {
"evalsmith-mcp": {
"command": "npx",
"args": [
"-y",
"evalsmith-mcp@0.1.0"
],
"env": {
"EVALSMITH_JUDGE_BASE_URL": "<YOUR_EVALSMITH_JUDGE_BASE_URL>",
"EVALSMITH_JUDGE_MODEL": "<YOUR_EVALSMITH_JUDGE_MODEL>",
"EVALSMITH_JUDGE_API_KEY": "<YOUR_EVALSMITH_JUDGE_API_KEY>",
"LOG_LEVEL": "<YOUR_LOG_LEVEL>"
}
}
}
}Cline
{
"mcpServers": {
"evalsmith-mcp": {
"command": "npx",
"args": [
"-y",
"evalsmith-mcp@0.1.0"
],
"env": {
"EVALSMITH_JUDGE_BASE_URL": "<YOUR_EVALSMITH_JUDGE_BASE_URL>",
"EVALSMITH_JUDGE_MODEL": "<YOUR_EVALSMITH_JUDGE_MODEL>",
"EVALSMITH_JUDGE_API_KEY": "<YOUR_EVALSMITH_JUDGE_API_KEY>",
"LOG_LEVEL": "<YOUR_LOG_LEVEL>"
}
}
}
}VS Code
{
"servers": {
"evalsmith-mcp": {
"type": "stdio",
"command": "npx",
"args": [
"-y",
"evalsmith-mcp@0.1.0"
],
"env": {
"EVALSMITH_JUDGE_BASE_URL": "<YOUR_EVALSMITH_JUDGE_BASE_URL>",
"EVALSMITH_JUDGE_MODEL": "<YOUR_EVALSMITH_JUDGE_MODEL>",
"EVALSMITH_JUDGE_API_KEY": "<YOUR_EVALSMITH_JUDGE_API_KEY>",
"LOG_LEVEL": "<YOUR_LOG_LEVEL>"
}
}
}
}Claude Desktop
{
"mcpServers": {
"evalsmith-mcp": {
"command": "npx",
"args": [
"-y",
"evalsmith-mcp@0.1.0"
],
"env": {
"EVALSMITH_JUDGE_BASE_URL": "<YOUR_EVALSMITH_JUDGE_BASE_URL>",
"EVALSMITH_JUDGE_MODEL": "<YOUR_EVALSMITH_JUDGE_MODEL>",
"EVALSMITH_JUDGE_API_KEY": "<YOUR_EVALSMITH_JUDGE_API_KEY>",
"LOG_LEVEL": "<YOUR_LOG_LEVEL>"
}
}
}
}Let your agent find its tools
Find servers like evalsmith-mcp from your agent. Add Koskamo once to search the index for endpoints, access requirements, and client setup. Public discovery needs no key.