PreMan
PreMan provides a simulation sandbox for engineering teams building AI agents that rely on MCP tools. The platform lets teams describe expected outcomes in plain English, run concurrent probes to test tool discovery and workflow decisions, and turn failures into reusable regression scenarios. It also offers a release-to-repair loop that connects git pushes to affected endpoints and generates reviewable fix pull requests.
- Developer Tools
- Artificial Intelligence
- AI Agents
- Software Only
Funding
Funding not disclosed
Founders
Founder details are not available yet.
Product
Problem
Teams building AI agents that rely on MCP tools face fragmented release processes where git history, test results, observability data, and customer impact information live in disconnected systems. This fragmentation makes it difficult to understand what changed in a release, which endpoints may behave differently, which customers or workflows are affected, and whether a proposed fix is actually safe.
Solution
PreMan provides a simulation sandbox that lets teams recreate production-like conditions for testing AI agents using MCP tools. Users describe expected outcomes in plain English, including paths agents should never take, and PreMan runs concurrent probes to check connection behavior, tool discovery, and workflow decisions. Failures are captured with their scenario, expected outcome, and evidence, becoming reusable regression scenarios that teams can rerun after any tool, model, or product change. The platform extends to a release-to-repair loop that connects git pushes to affected endpoints, targeted tests, customer-impact analysis, and reviewable fix pull requests, keeping production under watch after deployment.
Target Audience
Primary users are engineering teams building AI agents, MCP server developers, and product teams shipping agent-facing APIs who need to verify tool behavior, prevent regressions, and maintain production reliability.
Features
- Concurrent MCP discovery probes to verify tool availability and connection behavior
- Workflow simulation that captures tool-call evidence and decision paths agents take
- Reusable regression scenarios that preserve inputs, outputs, and failure evidence for rerunning after changes
- Plain-English rule definition for expected outcomes, including disallowed paths
- Production-like data and tool responses to recreate real conditions
- Desktop and terminal execution options for regression checks
- Public catalog of MCP servers for browser-based testing without installation
- Release-to-repair loop that builds affected-endpoint maps from commit diffs and generates reviewable fix pull requests with focused tests