MCP Dev Loop: Trello → Multi-Agent (Cursor, Claude Code, Windsurf, Copilot) → GitHub

Human-in-the-loop AI development — Trello tickets drive structured analysis, implementation, automated tests, commits, and GitHub Pull Requests across any AI coding agent.

❌ Not autonomous AI — the agent suggests, plans, and tests; you approve; the agent executes. ✅ Human-in-the-loop — no surprise commits, pushes, or unverified Trello updates.

Connect Trello via MCP and run a repeatable, safe dev loop across Claude Code, Cursor, Windsurf, VS Code Copilot, Claude Desktop, and Antigravity.

Architecture

┌──────────────┐     ┌───────────────────────┐     ┌────────────────────────────┐
│   Trello     │ ──► │   Trello MCP Server   │ ──► │ AI Coding Agent (Human Ctrl)│
│  (Tickets)   │     │ (Standalone / Node)   │     │ Claude Code / Cursor / etc.│
└──────────────┘     └───────────────────────┘     └─────────────┬──────────────┘
                                                                 │
                                ┌────────────────────────────────┼────────────────────────────────┐
                                ▼                                ▼                                ▼
                         ┌─────────────┐                  ┌─────────────┐                  ┌─────────────┐
                         │  Codebase   │ ───────────────► │   GitHub    │ ───────────────► │   Trello    │
                         │ Edits/Tests │   automated PR   │ PRs/Commits │   update/sync    │  comments   │
                         └─────────────┘                  └─────────────┘                  └─────────────┘

What's New in v2.0.0

🌐 Universal Cross-Agent Support: Native configuration templates for Claude Code, Cursor, Windsurf, VS Code Copilot, Claude Desktop, and Antigravity.
⚡ Standalone Source-Protected Runner: Packaged with esbuild into a single standalone minified bundle (dist/trello-mcp.cjs, ~349 KB) requiring zero node_modules installation.
🔍 Semantic List Auto-Detection (auto_detect_lists): Automatically maps "To Do", "In Progress", "Review", and "Done" columns on any board — no manual list ID setup.
🧪 Pre-Commit Test Runner (run_pre_commit_tests): Runs local test suites (npm test, pytest, cargo test, go test) with timeout guards before code is staged.
🚀 Automated GitHub Pull Requests (create_github_pr): Creates branches, opens GitHub PRs, and attaches links directly to Trello cards.
📦 Agensi.io Commercial Release Package: 1-command release packager (npm run package:agensi) producing a verified <50 MB distribution zip with root SKILL.md.

Supported AI Coding Agents

Agent	Scope	Setup Config	Command / Method
Cursor	Local / Global	`configs/cursor/mcp.json`	Settings → MCP → Add `trello`
Claude Code	Terminal CLI	`configs/claude-code/`	`claude mcp add trello node ./dist/trello-mcp.cjs`
Claude Desktop	macOS / Windows	`configs/claude-desktop/`	`%APPDATA%\Claude` or `~/Library/...`
Windsurf	Native Cascade	`configs/windsurf/mcp_config.json`	`~/.codeium/windsurf/mcp_config.json`
VS Code Copilot	Workspace	`configs/vscode-copilot/mcp.json`	`.vscode/mcp.json`
Antigravity	Agent SDK	`configs/antigravity/`	Native MCP stdio registration

Repository Structure

├── SKILL.md                          # Universal agent skill definition (Agensi.io compliant)
├── dist/
│   ├── trello-mcp.cjs                # Standalone compiled runner (349 KB)
│   └── mcp-dev-loop-agensi.zip       # Commercial release bundle (<110 KB)
├── configs/                          # Multi-agent config templates
│   ├── cursor/mcp.json
│   ├── claude-code/claude_code_config.json
│   ├── claude-desktop/claude_desktop_config.json
│   ├── windsurf/mcp_config.json
│   ├── vscode-copilot/mcp.json
│   └── antigravity/mcp_config.json
├── bin/
│   ├── setup.sh                      # 1-Click interactive setup (macOS / Linux)
│   └── setup.ps1                     # 1-Click interactive setup (Windows PowerShell)
├── docs/
│   ├── agensi-listing.md             # Commercial marketplace listing dossier & copy
│   ├── setup.md                      # Detailed manual setup guide
│   ├── safety-model.md               # Human-in-the-loop protection details
│   └── example-prompts.md            # Daily prompt patterns
├── scripts/
│   └── package-agensi.js             # Release packaging script
├── trello-mcp-server/                # Core engine source code
│   ├── server.js                     # Tool & prompt definitions
│   ├── lib/
│   │   ├── trelloClient.js           # Trello API + list auto-matcher
│   │   └── env.js                    # Multi-location credential loader
│   └── tests/                        # Automated unit tests
├── LICENSE                           # Open source MIT license (Copyright 2026 Mihretab Afework)
└── LICENSE-COMMERCIAL.md             # Commercial single-user license for Agensi buyers

Quick Start (1 Minute)

Option A: Interactive 1-Click Setup (Recommended)

# macOS / Linux
./bin/setup.sh

# Windows (PowerShell)
.\bin\setup.ps1

The setup script prompts for your Trello API credentials, tests the connection, and auto-configures your selected AI agent.

Option B: Manual Setup

Copy .env.example to .env and fill in:

TRELLO_API_KEY=your_trello_api_key
TRELLO_TOKEN=your_trello_token
TRELLO_BOARD_ID=optional_board_id

(Get credentials at trello.com/power-ups/admin)

Build standalone runner & test:
```
npm run build
npm test
```
Copy the appropriate configuration from configs/ to your agent's config directory.

Daily Workflow

Step	You say	Agent does
1. Pull	`analyze my assigned tasks`	Lists your assigned Trello queue
2. Plan	`analyze ticket <id>`	Full card extraction → PLAN (read-only, no code edits)
3. Approve	`proceed`	Gives approval to start implementation
4. Build	`implement ticket <id>`	Modifies codebase according to approved plan
5. Test	`run tests for ticket <id>`	Runs project test suite (`run_pre_commit_tests`)
6. Commit	`prepare commit for ticket <id>`	Shows diff + drafts ticket-linked commit message
7. PR	`create pr for ticket <id>`	Creates GitHub PR (`create_github_pr`) & links to Trello
8. Sync	`update trello ticket <id>`	Posts comment + moves card to Done / Review

MCP Tools Reference

Tool	Description
`get_my_cards`	List cards assigned to the authenticated user.
`get_card`	Complete extraction: description, comments, checklists, custom fields, attachments, activity.
`auto_detect_lists`	Semantically detects board columns (To Do, In Progress, Review, Done).
`run_pre_commit_tests`	Automatically detects and runs project tests before staging commits.
`create_github_pr`	Creates a git branch, opens a GitHub PR, and attaches the link to the card.
`mark_in_progress`	Moves card to In Progress (with semantic list resolution).
`mark_done`	Moves card to Done (with semantic list resolution).
`attach_commit`	Attaches commit or PR URL to card.
`add_comment`	Posts update comments to Trello.
`get_boards` / `get_board_cards` / `get_board_lists`	Board navigation tools.

Commercial Distribution on Agensi.io

To package this project for listing on Agensi.io:

npm run package:agensi

This compiles a standalone single-file bundle via esbuild, validates the root SKILL.md frontmatter, verifies package size (<50 MB, typically ~110 KB), and produces:

dist/mcp-dev-loop-agensi.zip

Complete seller copy, pricing ($14.00), and upload guide: docs/agensi-listing.md.

Safety & Human Oversight

Strict Approval Gates: AI never commits, pushes, or moves Trello cards without explicit user commands.
Pre-commit Test Guard: Verifies project test status before staging files.
Zero Secret Leaks: Automated scans skim diffs to ensure .env and API credentials are never committed.

Author & License

Author: Mihretab Afework (@Mih-Nig-Afe)
Open Source License: MIT License — Copyright (c) 2026 Mihretab Afework. See LICENSE.
Commercial Marketplace License: See LICENSE-COMMERCIAL.md.

MCP Dev Loop: Trello → Multi-Agent (Cursor, Claude Code, Windsurf, Copilot) → GitHub