evalsmith-mcp MCP Server
Evaluate RAG retrieval, agent tool calls and prompt regressions with deterministic metrics.
evalsmith-mcp scores 38 (Limited), licensed MIT and written in TypeScript. It installs as a local process. It does not declare tools yet. Filed under AI and memory. GitHub shows 1 stars. Install configs are generated for 10 clients. Being listed in the registry is not the same as being vendor-official.
Limited trust
38 of 100 from six public signals. Nobody can buy a point.
- OSI-style license: mit.
License, quality and maintenance
Where the 38 comes from
0 + 20 + 1 + 10 + 4 + 3 = 38. Each factor is capped, so no single signal can carry a listing.
Active in the official MCP registry, plus a known vendor namespace.
Registry: 0 of 30 points.How recently the repository saw a commit or release.
Maintenance: 20 of 20 points.GitHub stars and weekly downloads, on a log scale.
Popularity: 1 of 20 points.A clear, OSI-style open-source license.
License: 10 of 10 points.Not archived, a secure install path, a pinned version.
Security: 4 of 10 points.A real summary, a README with depth, listed tools.
Docs: 3 of 10 points.Install
Pick your client, copy the config, paste it in.
Needs Cursor or VS Code installed on this device. The button opens the app and asks you to confirm before it adds the server.
{ "mcpServers": { "diegoperezg7-evalsmith-mcp": { "command": "npx", "args": [ "-y", "evalsmith-mcp@0.1.0" ], "env": { "EVALSMITH_JUDGE_BASE_URL": "http://localhost:11434/v1", "EVALSMITH_JUDGE_MODEL": "llama3.1:8b", "EVALSMITH_JUDGE_API_KEY": "YOUR_EVALSMITH_JUDGE_API_KEY", "LOG_LEVEL": "info" } } }}
Environment variables
| Name | Required | Notes |
|---|---|---|
EVALSMITH_JUDGE_BASE_URL | Optional | OpenAI-compatible endpoint for the optional LLM judge (default: Ollama on localhost) |
EVALSMITH_JUDGE_MODEL | Optional | Model used by the optional LLM judge |
EVALSMITH_JUDGE_API_KEYSecret | Optional | API key for the judge endpoint, if it needs one |
LOG_LEVEL | Optional | pino log level; logs go to stderr |
Tools
All capabilitiesTools show up when the registry metadata or the repository README lists them. Public READMEs are read directly. A GitHub token is used only when one is already configured.
About
MCP server and CLI to evaluate RAG retrieval, agent tool calls and prompt regressions, with CI gates. No paid keys.
Frequently asked questions
What is evalsmith-mcp MCP Server?
Evaluate RAG retrieval, agent tool calls and prompt regressions with deterministic metrics.
How do I install evalsmith-mcp MCP Server in Claude Desktop, Cursor, or VS Code?
Pick your client in the Install section on this page and copy the generated config into the client's config file. Step-by-step guides: Claude Desktop, Cursor, and VS Code. evalsmith-mcp MCP Server runs locally from the npm package evalsmith-mcp with the command npx -y evalsmith-mcp@0.1.0.
Is evalsmith-mcp MCP Server free and open source?
Yes. evalsmith-mcp MCP Server is released under the MIT license, an OSI-approved open-source license, so you can use, modify, and self-host it under those terms.
Does evalsmith-mcp MCP Server support remote (HTTP) connections?
No remote endpoint is listed for evalsmith-mcp MCP Server. It runs locally on your machine over stdio, started by your MCP client.
How is evalsmith-mcp MCP Server's trust score calculated?
evalsmith-mcp MCP Server scores 38 out of 100 (Limited). The score adds six capped public signals: registry status, maintenance, popularity, license, documentation, and security. Nobody can pay for points. Read the methodology for the full formula.
Reviews
No reviews yet—Be the first to review
No published reviews yet. Used it? Tell the next person how it went.
Sign inA verified email is required to review this server.
Add badge to your README
Paste either snippet into the repository README. The image links to this listing.