Releases·Tool Release
Dev open-sources Kaktoos, an MCP tool that verifies AI-written API integrations against OpenAPI contracts
A developer introduced Kaktoos, an open-source tool addressing a gap in AI coding agents: when an agent both writes an API integration and its tests, passing tests may not confirm the integration matches the real API. The tool targets verification of AI-written integrations against actual API behavior. The post asks the community for technical feedback.
5 engagement1 sourceThu, Sep 10, 2026, 03:57 PM
AI Overview
- Kaktoos runs AI-agent-written API integration code against the OpenAPI contract and the real API, returning structured failures for the agent to fix
- The tool executes multi-step API workflows and flags missing required fields, wrong types, and unexpected status codes or content types
- Kaktoos exposes verification via MCP for coding agents and runs the same checks in GitHub Actions
- The Reddit post (KaktoosLabs/kaktoos, Sept 10, 2026) asks practitioners whether independent verification catches problems AI-written tests miss and if it fits CI
Entities
Related
Products·building
AI agent sandbox that verifies API integrations end-to-end before production
A playground that forces AI agents (like Cursor, Claude Code) to prove an API integration works by running the full workflow in a sandbox and getting a receipt, rather than relying on tests. It includes two tasks on a Descope integration—one normal, one with a planted bug—to test if the agent finds it. Aimed at developers who want to avoid broken production integrations.
A playground that forces AI agents (like Cursor, Claude Code) to prove an API integration works by running the full workflow in a sandbox and getting a receipt, rather than relying on tests. It includes two tasks on a Descope integration—one normal, one with a planted bug—to test if the agent finds it. Aimed at developers who want to avoid broken production integrations.
4 engagement1 sourceredditSun, Jul 26, 2026, 01:48 AM
Products·building
Full-stack AI coding agent that builds and deploys apps from natural language
This is an AI coding agent that converts natural language requirements into full-stack applications, generating frontend, backend, database structures, and APIs. It iterates through conversations to test and deploy working applications, targeting SaaS products and internal business tools. The project is currently in development and seeking feedback from experienced builders.
This is an AI coding agent that converts natural language requirements into full-stack applications, generating frontend, backend, database structures, and APIs. It iterates through conversations to test and deploy working applications, targeting SaaS products and internal business tools. The project is currently in development and seeking feedback from experienced builders.
0 engagement1 sourceredditSat, Aug 8, 2026, 06:58 AM
Products·building
Orinyx — open-source MCP for independently verifying AI agent outputs
Orinyx is an open-source MCP server that checks AI agent outputs against defined requirements and supporting evidence instead of trusting the agent's own response. It targets developers of agent workflows, starting with healthcare-related evaluations. First release, Orinyx MCP R1, is scheduled for November 1.
Orinyx is an open-source MCP server that checks AI agent outputs against defined requirements and supporting evidence instead of trusting the agent's own response. It targets developers of agent workflows, starting with healthcare-related evaluations. First release, Orinyx MCP R1, is scheduled for November 1.
1 engagement1 sourceredditWed, Oct 14, 2026, 01:48 PM
Products·building
OpenPitStop: independent referee for AI coding agents
OpenPitStop is an open-source tool that acts as an external referee for AI coding agents, independently checking their work to verify fixes actually hold. It addresses the problem of trusting AI agents by providing a verification layer outside the agent. Built for developers using AI coding agents, it was tested on a broken MiniShop application.
OpenPitStop is an open-source tool that acts as an external referee for AI coding agents, independently checking their work to verify fixes actually hold. It addresses the problem of trusting AI agents by providing a verification layer outside the agent. Built for developers using AI coding agents, it was tested on a broken MiniShop application.
32 engagement1 sourceredditMon, Aug 24, 2026, 07:55 AM
Products·building
AI API reverse engineering tool that captures traffic and generates tested client code
A tool that lets developers capture API traffic by browsing in their browser (or having an AI agent do it), then automatically generates a working API client. An AI agent tests the generated code against the real API. It solves the problem of discovering and reverse-engineering undocumented APIs for developers.
A tool that lets developers capture API traffic by browsing in their browser (or having an AI agent do it), then automatically generates a working API client. An AI agent tests the generated code against the real API. It solves the problem of discovering and reverse-engineering undocumented APIs for developers.
2 engagement1 sourceredditSat, Jul 25, 2026, 10:09 PM