JSPM

  • ESM via JSPM
  • ES Module Entrypoint
  • Export Map
  • Keywords
  • License
  • Repository URL
  • TypeScript Types
  • README
  • Created
  • Published
  • Downloads 89
  • Score
    100M100P100Q98586F
  • License MIT

Stops the Claude Code correction loop — maps dependencies before execution, verifies completeness after.

Package Exports

  • united-agents-mcp
  • united-agents-mcp/dist/index.js

This package does not declare an exports field, so the exports above have been automatically detected and optimized by JSPM instead. If any package subpath is missing, it is recommended to post an issue to the original package (united-agents-mcp) to support the "exports" field. If that is not possible, create a JSPM override to customize the exports field for this package.

Readme

United Agents MCP

Stop telling Claude "you didn't finish." United Agents enforces complete task execution automatically.

The Problem

You ask Claude Code to fix something. It touches one file, says "done." But the bug required changes in 4 connected files. You say "you didn't finish." Claude fixes another file, says "done." Still broken. You repeat this 3-5 times, burning tokens on every round.

This happens because Claude doesn't map the full dependency chain before starting — and declares success after the first file it touches.

The Solution

United Agents is an MCP server that:

  1. Maps every connected file before Claude starts (API routes, DB layers, components)
  2. Blocks premature "done" — Claude cannot report completion until all files are addressed
  3. Forces autonomous continuation — if incomplete, Claude keeps working without asking you

The correction loop stops. Tasks complete in one round.

Install

npm install -g united-agents-mcp

Setup (one time)

Add to Claude Code:

claude mcp add united-agents node $(which united-agents-mcp) --scope user

Set up your project (run once per project in Claude Code):

Use the setup_project tool with project_root "/your/project/path"

This writes a CLAUDE.md into your project with the enforcement rules. After this — you're done. The loop runs automatically on every task.

How It Works

You give Claude a task
        ↓
Claude calls trace_dependencies → maps all connected files
        ↓
Claude works through every file in the map
        ↓
Claude calls verify_completeness before saying "done"
        ↓
🚫 INCOMPLETE → Claude continues automatically (no asking you)
        ↓
✅ COMPLETE → Claude reports done
        ↓
You never had to say "you didn't finish"

Tools

trace_dependencies

Maps every file connected to the one you want to change.

file: "LeadPipeline.tsx"
project_root: "/Users/you/your-project"

Returns:

  • 🎯 Direct files — must change
  • 🌐 API routes — called by the file
  • 🗄️ DB layers — Supabase/Prisma connections
  • 🔗 Connected imports — shared utilities

verify_completeness

Checks git diff against the dependency map. If anything is missing — Claude continues automatically.

setup_project

Writes CLAUDE.md enforcement rules into your project. Run once per project.

Stack Support

Currently optimized for:

  • Next.js + TypeScript
  • Supabase
  • Vercel deployments

More stacks coming soon.

Real Example

Before United Agents — fixing a drag bug in KanbanPipeline.tsx:

  • Round 1: Claude fixes component, says done. Bug persists.
  • Round 2: Claude fixes event handler, says done. Bug persists.
  • Round 3: Claude finds DB layer issue, says done. Bug persists.
  • Round 4: Claude fixes RLS policy, says done. Finally works.
  • 4 rounds. 4x token cost. 45 minutes wasted.

After United Agents:

  • trace_dependencies maps 7 connected files immediately
  • Claude works through all 7 in one pass
  • verify_completeness confirms complete
  • 1 round. Done.

License

MIT

Author

Nursultan Orynbassar — unitedagents.dev