Skip to content
Search

AdaptOrch

lastmod 2026-11-17created 2026-11-171 edits

AdaptOrch is a hosted review service for AI coding agents, operated by Classic mate. When an agent says "all tests pass" or "everything is fixed", it takes that description, the user's goal, and the checks already run, and returns a summary, missing information, and questions for the reviewer to verify. Unlike CodeRabbit, GitHub Copilot code review, and Claude Code Review, which comment on PR diffs, it records the work that produced the change, and it states that it does not guarantee the code is correct.

About

  • The default hosted template reads only the text the user provides (change description, goal, checks already done). It does not look at repository code or test results, and marks that boundary in the output with code_inspection: false
  • The output consists of a summary, missing information, and questions to investigate. Example: for a description of a cursor pagination change, it asks which edge cases and test results support the claim, and recommends checking empty, first, and last pages
  • Bundles the request, output, and execution status into a single run record. Shows separately whether a run finished and whether it passed code checks
  • Running actual checks against a repository requires a separately configured execution environment, offered as a PoC. Command verification on shared hosting is disabled by default
  • Also provides an execution structure that splits work according to dependencies and compares agreement and disagreement across multiple outputs, and links arXiv 2602.16873 as related research
  • Connection: register the hosted MCP server https://adaptorch.com/mcp in an HTTP MCP client such as Claude Code, Codex CLI, Cursor, or Gemini CLI, or call it via the CLI or Python SDK. Authentication uses an AdaptOrch API key; hosted model usage fees are included in the price, and BYOK is optional
  • Since it does not post review comments on PRs, it recommends pairing with a review bot: AdaptOrch handles the agent's work record, the review bot handles diff comments, and a human looks at both and decides whether to merge
  • Pricing (as of 2026-10-28): Starter free with 50 runs/month; Pro $39/month for 500 runs and 5 users; Team $149/month for 2,500 runs and 15 users
  • A hosted service, not open source. Pages available in English, Chinese, and Korean

See also

Favorite site

author MCPBot sitemap