Callout is a multi-perspective architecture review tool that works as an MCP server and can be embedded in mainstream AI programming environments such as Claude Code, Cursor, VS Code, and Windsurf. It is designed to solve a common problem: although AI programming tools can greatly increase the speed of writing code, few people check whether developers are building the right thing. Callout is positioned as a "second pair of eyes," using 9 expert perspectives to catch issues developers may overlook.
Purpose and Background Callout's core purpose is to provide multi-perspective architecture and decision review during AI-assisted programming. Official materials mention a real case: a product was built in 6 days using Claude Code, including 186 files and 691 tests; a multi-perspective review on day 5 found that 60% of the content was overengineering, and the originally planned 9 days of work was reduced to 3.5 days. The key insight is: the problem is not poor code quality, but wrong priorities and missed blind spots. Callout automates this review capability and provides it to every developer.
Main Capabilities
- review: Reviews a project from 9 expert perspectives, divided into 3 groups. Each finding is classified as MUST FIX, SHOULD FIX, or GOOD, with effort estimates attached, beginning with an executive summary.
- coach: An AI collaboration blind spot detector that analyzes project setup, development habits, and knowledge gaps, pointing out issues you are not aware of when collaborating with AI programming tools.
- idea_score: A quantitative idea validator that scores ideas across 10 dimensions (market size, feasibility, moat, revenue potential, etc.), defaults to a skeptical stance, and returns a conclusion of CONTINUE, SIMPLIFY, PAUSE, or DELETE.
- Domain Knowledge Guard: Binds official documentation to code paths. After binding, AI must read the referenced documentation before modifying protected files. Related tools include doc (register/remove/list/check), doc_gate, and renovation.
- test_translate: Parses test output into plain language, explaining what was tested and what failed, and generates manual test scripts.
- recommend: Detects what a project needs (authentication, database, payments, etc.), recommends the best tools, including installation commands, setup steps, and alternatives.
How to Use Callout requires no global installation. Run the 30-second install command in the project directory, and it will automatically detect the editor and configure the MCP server. After installation, restart the editor (or re-enter Claude Code in the project directory), then invoke features through natural language instructions, for example saying "review" for review, "coach" for collaborative blind spot analysis, and "idea_score" for idea scoring. In multi-project scenarios, you can say "portfolio" to get cross-project health status and resource allocation suggestions. Manual configuration applies to Cursor/VS Code (add to .cursor/mcp.json or .vscode/mcp.json) and Windsurf (add to ~/.codeium/windsurf/mcp_config.json), with the same format.
Callout also supports intelligent automatic triggering: after running init, describing a new feature will first be discussed before coding; opening a new project will automatically perform coach analysis; modifying files protected by documentation will be intercepted until the documentation is read; completing a milestone will suggest a review.
Use cases Suitable for individuals or teams developing with AI programming tools, especially for scenarios that require controlling architectural direction, avoiding over-engineering, validating product ideas, ensuring AI follows documentation, and obtaining tool selection advice before coding.
Limitations and notes Callout itself is an MCP server and does not call an LLM. It collects project context (file structure, dependencies, README, git history) and assembles prompts from an expert perspective, and the review is executed by the AI tool the user already has. Therefore, it has zero API cost, does not require an API key, and is compatible with the model the user is currently using. It can integrate with optional companion MCP servers, and no additional configuration is required after installation. Regarding charging, pricing, platform support details, and specific effects, no further information is provided in the materials.
