Promptfoo

Package

Local LLM evaluations and red-team tests with an official MCP server

Price
Free CLI; model calls billed separately
Access
No account; model credentials as needed

About

CLI and library for comparing prompts, models and RAG systems with datasets and assertions, plus adversarial testing of AI applications. Evaluations run locally or in CI; model calls can incur provider charges. Its official MCP server exposes evaluation and red-team tools to agents.

What you can do with it

  • Compare prompts and model outputs against datasets and assertions
  • Probe an AI application with adversarial red-team test cases
  • Let an agent validate configurations and run evaluations through local MCP

Get started

  1. Install Node.js and configure prompts, providers and test cases
  2. Provide credentials for model providers used in the evaluation
  3. Run evaluations or expose tools with the local MCP command

Example

npx promptfoo@latest validate
# Requires a configured evaluation project and any provider credentials
npx promptfoo@latest eval

Details

Hosting
Runs locally, Self-hosted
Available in
Worldwide
Official SDKs
JavaScript/TypeScript
MCP server
Local

Tasks

Alternatives

Other tools for the same tasks.

Last checked on .