hn.today

Show HN: MCPJam - the first testing & evaluations platform for MCP servers

mcpjam.com11 points6 comments

MCPJam is a testing and evaluation platform that exercises MCP servers across real AI clients (ChatGPT, Claude, Copilot, Cursor, VS Code and others) to reveal where behavior diverges and why. It provides an Inspector to run prompts simultaneously in multiple clients, an OAuth & EMA debugger to pinpoint auth breakdowns, and cross-client traces that show tool calls and UI differences. Tests can run locally via a CLI/Inspector, in code via an SDK, or in the web/desktop app. Simulated “swarms” run agent personas through user journeys across chosen clients and servers, while a user-testing mode invites real people into sandboxed chats to measure sentiment and usability. Core components (Inspector, CLI, SDK, local evals and conformance checks) are open source; enterprise features include SSO, audit logs, and a DPA, and MCPJam asserts it never touches live traffic in pre-production usage.

The platform integrates into CI/CD to gate merges: suites score goal completion, tool choice and arguments across repeated runs so non-deterministic behavior is evaluated rather than brittle-asserted. Example suite outputs show run-by-run health (e.g., Payments operations averaging 98% across 12 runs) with per-run metrics like latency, tokens, calls, and pass rates; failing checks block PRs until MCPJam evals meet set thresholds. Target users include developers, product and engineering managers, AI/platform leads and enterprises that need assurance that MCP servers behave correctly across every major client.

Read on mcpjam.com6 comments on Hacker News

Summary generated by AI from the linked article. hn.today is not affiliated with Hacker News or Y Combinator.

More in Programming

The daily digest

Today's best Hacker News stories, summarized and screenshotted, one email a day.