MCP: Using Your Favorite AI Assistant with DoneIsGood

1. The Why: The Power and Peril of External AI Agents

Modern AI tools—whether it's Claude Desktop, Cursor IDE, Claude Code CLI, ChatGPT, or Google Gemini—are extraordinary reasoning engines. However, in isolation, they suffer from two major shortcomings:

1. Context Blindness: They don't know what you committed to yesterday, which quadrant your priorities sit in, or what meetings you have this afternoon. You waste endless prompt cycles re-explaining your context.
2. The Security Dilemma: Typical integrations force you to hand over unlimited, un-scoped database access. A rogue agent or accidental hallucination could delete shared company boards or overwrite client project records.

DoneIsGood solves this by providing a high-performance Model Context Protocol (MCP) Server coupled with Native AI Governance Guardrails built directly into the UI.

2. The DoneIsGood Blueprint: Governance & Safety Architecture

Unlike standard API keys with wide-open permissions, DoneIsGood enforces strict multi-layer boundaries on external agents:

3. Step-by-Step How-To in DoneIsGood

Step 1: Generate Your Personal AI Access Token

1. Open DoneIsGood and navigate to Settings $\rightarrow$ AI & Integrations (or the MCP Configuration panel).
2. Click Generate New MCP Token.
3. Label your token clearly (e.g. "Claude Desktop - MacBook Pro").
4. Copy the secure secret key immediately. (For security, tokens are encrypted and cannot be viewed again).

Step 2: Configure Granular Matrix Access

Before connecting an agent, define its boundaries:

• Set sensitive boards to Private (Hidden) so agents cannot even see their existence.
• Set team boards to Read-Only (to reference collective tasks without modifications) or Read & Write (Assigned Tasks Only) (to let your agent update subtask steps, log focus time, and mark completion on tasks assigned to you).
• Designate an AI Default Matrix (e.g. "AI Inbox" or your personal triage board) where new agent-created tasks will land.

Step 3: Connect Your Favorite Client

Option A: Claude Desktop (claude_desktop_config.json)

Add the following snippet to your Claude Desktop configuration file:

{
  "mcpServers": {
    "doneisgood": {
      "command": "npx",
      "args": [
        "-y",
        "@doneisgood/mcp-server",
        "--token",
        "YOUR_PERSONAL_ACCESS_TOKEN"
      ]
    }
  }
}

Option B: Cursor IDE (.cursor/mcp.json)

Inside your project or global settings, add:

{
  "mcpServers": {
    "doneisgood": {
      "command": "npx",
      "args": ["-y", "@doneisgood/mcp-server"],
      "env": {
        "MANAMANAH_API_TOKEN": "YOUR_PERSONAL_ACCESS_TOKEN"
      }
    }
  }
}

Option C: ChatGPT Custom Actions or Gemini Spark

DoneIsGood provides an OpenAPI 3.1 specification at https://doneisgood.com/openapi.json. Import the schema into ChatGPT Custom GPTs with Bearer Authentication using your personal access token.

4. Practical Agent Commands (What You Can Do)

Once connected, you can interact with DoneIsGood conversationally inside your favorite AI interface:

• "What are my top Q1 urgent tasks in DoneIsGood today?"
• "Based on our conversation about the database migration, create a Q2 task with 4 checklist steps and assign it to my Engineering matrix."
• "Create a reference document in DoneIsGood titled 'JIRA PROJ-452: Authentication Architecture Spec' with the design requirements and tagged 'jira', 'spec'."
• "Append the latest meeting notes and action items to my 'Q4 Product Roadmap' document in DoneIsGood."
• "Search my DoneIsGood documents for any existing notes about 'OAuth token refresh'."
• "Log 1.25 hours of deep work against the 'API Refactor' task in DoneIsGood."
• "What habits have I completed today, and what is my current streak?"
• "Create a watchlist of "The Gentlemen" on Netflix with a short synopsis of each episode as steps in DoneIsGood"

Every task and document created by an external agent automatically receives the createdByAgent tag and appears in your in-app audit trail.

5. Security Guardrails & Emergency Controls

• 1-Click Emergency Kill Switch: If an external agent misbehaves or goes into an unexpected loop, open DoneIsGood, click the top Security / Shield icon, and toggle Emergency AI Pause. This instantly rejects all incoming MCP requests at the server level without invalidating your client configuration files.
• In-App AI Agent Audit Trail: Review the real-time Audit Log under Settings. It tracks every read, write, and completion request, recording the timestamp, agent name, IP address, and payload parameters (with sensitive auth tokens automatically redacted).
• Shared Board Governance: On shared matrices granted Read & Write access, external agents can create new tasks (which are automatically assigned to you) and manage checklist steps, personal scheduling, and completion states on tasks assigned to you. Other teammates' tasks remain protected and can never be modified or leaked.

6. Next Steps & Related Guides

• Native AI vs. External MCP Agents: Learn when to use DoneIsGood's built-in voice coach vs external LLM agents.
• Mastering the Eisenhower Matrix: Ensure external agents structure tasks according to Eisenhower principles.
• Project Management in DoneIsGood: Coordinate AI-assisted workflows across shared team boards.