Skip to main content
Glama

Callout

npm version npm downloads License: MIT

You used AI to write code for 3 days. Callout tells you 60% of it was over-engineered.

AI coding tools make you 10x faster at writing code. Nobody checks if you're building the right thing. Callout is the second pair of eyes — 9 expert perspectives that catch what you miss, delivered as an MCP server that works inside Claude Code, Cursor, VS Code, and Windsurf.


Why Callout exists

A real product was built using Claude Code in 6 days: 186 files, 691 tests. A multi-perspective review on Day 5 found that 60% was over-engineered. 9 days of planned work reduced to 3.5 days. The key insight: the problem wasn't bad code — it was wrong priorities and missed blind spots.

Callout makes that review available to every developer, automatically.


Related MCP server: checkyourself

30-second setup

npx callout-dev setup

No global install needed. Run this from your project directory. Callout auto-detects your editor and configures the MCP server.

After setup, restart your editor (or re-enter Claude Code from the project directory), then say:

Review this project

Claude Code:

claude mcp add callout -- npx callout-dev

Cursor / VS Code — add to your editor's MCP config (.cursor/mcp.json or .vscode/mcp.json):

{
  "mcpServers": {
    "callout": {
      "command": "npx",
      "args": ["callout-dev"]
    }
  }
}

Windsurf — add to ~/.codeium/windsurf/mcp_config.json with the same format.

After any manual setup, restart your editor so it picks up the new MCP server.


Quick Start: What to say

Phase

What you want

Say this to AI

What happens

Get started

Check my AI collaboration habits

"Coach me"

Scans your project setup, finds blind spots, outputs improvement checklist

Evaluate this project

"Review this project"

9 expert perspectives review your project, categorize as MUST FIX / SHOULD FIX / GOOD

Is this idea worth building?

"Score this idea"

10-dimension scoring, verdict: CONTINUE / SIMPLIFY / PAUSE / DELETE

Go deeper

Get advice on a specific decision

"Review focus: should I use Supabase?"

Multi-perspective review focused on one question

Make AI read docs before coding

"Bind Stripe docs to payment code"

AI must read official docs before modifying protected files — no more guessing

What tool should I use?

"I need to add payments"

Detects project needs, recommends best tool with install steps

Manage

See all findings and progress

"Show my todos"

All review/coach findings auto-collected, sorted by priority

Multi-project? Say "portfolio" for cross-project health status and resource allocation advice.


Core Features

review — 9 expert perspectives on your project

3 groups, 9 perspectives:

Group

Perspectives

What they catch

Technical

CTO, Security, DevOps

Over-engineering, vulnerabilities, deployment gaps

Business

Product, Customer, Strategy

Missing features, UX gaps, business model issues

Founder

Investor, Unicorn Founder, Solo Entrepreneur

Market fit, scalability, solo feasibility

Every finding is categorized as MUST FIX, SHOULD FIX, or GOOD with effort estimates. Starts with an Executive Summary.

> Review this project                              — full 9-perspective review
> Review with founder group                        — investor + unicorn founder + solo entrepreneur
> Review focus: should I use Supabase or Firebase?  — focused on one decision
> Review security + CTO only                       — selected perspectives

coach — AI collaboration blind spot detector

Analyzes your project setup, development habits, and knowledge gaps. Tells you what you don't know you're doing wrong when working with AI coding tools.

idea_score — Quantitative idea validator

Scores your idea across 10 dimensions (market size, feasibility, moat, revenue potential, etc.) with a skeptical default stance. Returns a verdict: CONTINUE, SIMPLIFY, PAUSE, or DELETE.

Domain Knowledge Guard — Make AI read docs before coding

Bind official documentation to code paths. Once bound, AI must read the referenced docs before modifying protected files.

> Bind Stripe docs to payment code     — register a doc binding
> Check if my code matches the docs    — compliance verification
> Start a renovation for stripe        — systematic domain knowledge audit

Tools: doc (register/remove/list/check), doc_gate, renovation

test_translate — Plain-language test results

Parses test output and tells you: what's tested, what failed, and produces a manual test script.

recommend — Tool recommendations

Detects what your project needs (auth, database, payments, etc.) and recommends the best tool — with install commands, setup steps, and alternatives.

Smart auto-triggers

After running init, Callout speaks up automatically:

  • New feature described → discusses before coding

  • Opening a new project → coach analysis

  • Modifying doc-protected files → gates until docs are read

  • Completing a milestone → suggests review


Works Best With (optional companion MCPs)

Callout integrates with these MCP servers when installed — no configuration needed:

MCP

What it adds

Install

Context7

Auto-pulls latest docs for 9000+ public libraries — feed into doc register

claude mcp add context7 -- npx -y @upstash/context7-mcp

Sequential Thinking

Structured step-by-step reasoning for complex renovations and multi-perspective reviews

claude mcp add sequential-thinking -- npx -y @anthropic-ai/mcp-server-sequential-thinking

Firecrawl

Scrapes any web docs into markdown — bind with doc register

claude mcp add firecrawl -- npx -y firecrawl-mcp


How it works

Callout is an MCP server. It collects your project context (file structure, dependencies, README, git history) and assembles expert-perspective prompts. Your AI tool executes the review.

Callout never calls an LLM itself. Zero API cost. No API key needed. Works with whatever model you already use.


All 19 tools

Tool

Description

review

9-perspective architecture review (3 groups: technical, business, founder)

coach

AI collaboration blind spot detector

idea_score

10-dimension idea scoring with verdict

test_translate

Turn test output into plain-language test plan

doc

Domain Knowledge Guard — register/remove/list/check doc bindings

doc_gate

Gate check: returns docs AI must read before modifying a file

renovation

Systematic domain knowledge audit against official docs

recommend

Detect project needs and recommend best tools

recommend_dismiss

Dismiss a recommendation scenario

recommend_reset

Reset all dismissed recommendations

todo_add

Add a finding to the project todo list

todo_update

Update todo status

todo_list

View todos by priority or status

todo_summary

Project health overview

init

Initialize auto-trigger rules

callout_help

Show usage guide

save_review_findings

Save review summary for progress tracking

set_target_user

Set target user for customer perspective

portfolio

Multi-project overview with resource allocation advice


Contributing

Contributions welcome. Please open an issue before submitting a PR for significant changes.

npm install
npm run build
npm test

License

MIT

Available Tools

16 tools
callout_helpA

Show what Callout can do and how to use it. Call this when a user first connects or asks about Callout.

Inputs: no named parameters.

coachA

AI collaboration coach. Analyzes your project setup, development habits, and knowledge blind spots to reveal what you don't know you're doing wrong when working with AI coding tools. Does NOT review code quality (use review for that) — instead reviews YOUR behavior patterns.

Inputs: project_path.

idea_scoreA

Score your idea across 10 dimensions with a skeptical default stance. Produces quantitative scores (1-10) for market size, feasibility, moat, revenue, time-to-market, user validation, resource efficiency, scalability, founder fit, and risk/reward. Returns a verdict: CONTINUE, SIMPLIFY, PAUSE, or DELETE.

Inputs: idea_description, project_path.

initA

Initialize Callout in a project. Returns CLAUDE.md rules that enable automatic review/coach/recommend triggers. The host should append these rules to the project's CLAUDE.md file.

Inputs: project_path.

portfolioB

Multi-project overview for founders managing multiple projects. Scans a directory for all projects, shows health status (todos, review history, risks), and asks AI to give resource allocation advice from a founder perspective.

Inputs: projects_root.

recommendB

Detect what the project needs and recommend the best tool/service for each scenario. Recommends auth, database, payments, deployment tools etc. based on current task and existing dependencies. Same scenario is not recommended twice.

Inputs: task, project_path.

recommend_dismissA

Dismiss a recommendation scenario so it won't be suggested again. Use when the user says they don't need a recommendation for a particular area.

Inputs: scene (required), project_path.

recommend_resetA

Reset all dismissed recommendation scenarios. After reset, recommend will detect and suggest tools again for all scenarios. Also shows what was previously dismissed.

Inputs: project_path.

reviewA

Multi-perspective architecture review. Analyzes full project context and produces actionable findings from expert viewpoints. Use focus parameter to zoom in on a specific feature, page, or decision. Use perspective_group to select a predefined group (technical/business/founder) instead of listing individual perspectives.

Inputs: perspectives, perspective_group, focus, customer_role, +1 more.

save_review_findingsA

Save a summary of review findings to the project history. Call this after completing a review to enable progress tracking across reviews.

Inputs: summary (required), project_path.

set_target_userA

Set who the target user of this project is. This is used by the customer perspective in reviews to give feedback as that specific type of user. Auto-detected from README/CLAUDE.md if not set manually.

Inputs: target_user (required), project_path.

test_translateA

Translate test results into plain language. Tells a non-technical founder what the tests cover, what failed, and exactly what still needs manual verification. Produces a 15-minute manual test script.

Inputs: test_output, project_path.

todo_addB

Add an item to the project todo list. All findings from review and coach should be added here.

Inputs: title (required), priority (required), description, source, +1 more.

todo_listC

Show all todo items, optionally filtered by status or priority.

Inputs: status, priority, project_path.

todo_summaryB

Quick health check: how many items open, in progress, done. Shows top priority items.

Inputs: project_path.

todo_updateC

Update a todo item status or details.

Inputs: id (required), status, title, description, +2 more.

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    TDQS

    A3.7/5.0

    Scored across 16 tools

    Disambiguation4/5

    Most tools have distinct purposes, but there is some overlap between 'coach' and 'review' that could cause confusion, as both analyze aspects of the project. However, their descriptions clarify that 'coach' focuses on developer behavior patterns while 'review' focuses on architecture and expert perspectives, which helps mitigate misselection.

    Naming Consistency5/5

    Tool names follow a consistent snake_case pattern with clear verb_noun structures, such as 'todo_add', 'todo_list', 'save_review_findings', and 'set_target_user'. This predictability makes the tool set easy to navigate and understand.

    Tool Count5/5

    With 16 tools, the server is well-scoped for its purpose of project management and AI collaboration support. Each tool serves a specific function, such as idea scoring, reviews, recommendations, and todo management, without feeling excessive or lacking in coverage.

    Completeness5/5

    The tool set provides comprehensive coverage for the domain, including project initialization, idea evaluation, multi-perspective reviews, recommendation management, and todo tracking. There are no obvious gaps; tools like 'recommend_dismiss' and 'recommend_reset' add nuance, ensuring agents can handle full workflows without dead ends.

    Maintenance

    ActivityInactive
    ResponsivenessOutdated data

    Related MCP Connectors

    Related MCP Servers

    • A
      license
      Not graded
      quality
      D
      maintenance
      An MCP server that provides AI-powered code review and architecture analysis, simulating the perspective of an experienced staff engineer. It integrates with IDEs to review diffs, design decisions, and tradeoffs through natural language.
      1
      MIT
    • A
      license
      Not graded
      quality
      D
      maintenance
      An MCP server that equips AI assistants with specialized, opinionated development playbooks for building microservices, frontends, databases, and DevOps pipelines. It integrates with clients like VS Code Copilot, Claude Desktop, and Cursor to provide battle-tested expertise.
      36 npm
      MIT