2026 Gov AI Subsidy  ·  Up to NT$20M  ·  We handle everything   Free Consultation →

MCP Connection Guide

Connect Your AI Assistant to the Squid Project Management System

Use the open Model Context Protocol standard to connect any MCP-compatible AI assistant to manage projects, requirements, and releases directly.

On This Page
What is MCP?
Service Endpoints
Auth Architecture
Available Tools
Setup Steps
Usage Examples
Tech Specs
FAQ

What is MCP?

Model Context Protocol (MCP) is an open standard developed by Anthropic that allows any MCP-compatible AI assistant to securely access external systems. Through MCP, an AI assistant can interact directly with the Squid API to manage projects, track requirements, and schedule plans — not tied to any specific vendor or platform.

Before You Begin

①
Nautilus Account

You need a Nautilus account to log in. If you don't have one, register at Nautilus.

②
MCP Client

Any MCP-compatible client, such as Claude Desktop, Claude Code, or any third-party implementation.

Service Endpoints

Endpoint URL
MCP Server https://mcp-server-production-f4cuesozga-de.a.run.app/mcp
Health Check https://mcp-server-production-f4cuesozga-de.a.run.app/health
OAuth2 Discovery https://mcp-server-production-f4cuesozga-de.a.run.app/.well-known/oauth-authorization-server

Authentication Architecture

MCP Server authenticates via Nautilus OAuth2 — the entire flow is handled automatically.

💬
MCP Client
Claude Desktop / Code
①→
🔌
MCP Server
Discovery + Proxy
②→
🔐
Nautilus
Authorization Server
③→
🦑
Squid API
Resource Server
① Discovery

MCP Client requests OAuth2 Metadata from MCP Server to locate the Nautilus authorization endpoint.

② User Login

Browser opens the Nautilus login page. After the user authorizes with their Nautilus account, a JWT Token is issued.

③ API Calls

MCP Server proxies Squid API calls using the Token, with automatic background refresh — no manual re-login needed.

43 Available Tools

Once connected, Claude can use the following tools to manage your Squid system.

Requirement Management 8

Create, query, update, delete, change status, list, submit validation reports, and associate with plans.

Project Management 9

Create, query, update, delete, archive, change status, list, track progress, and associate with plans.

Category Management 6

Create, query, update, delete, disable, and list categories.

Plan Management 9

Create, query, update, delete, manage status, partial updates, create versions, list versions, and associate requirements.

Proposal Management 5

Create, query, view details, create versions, and list versions for project proposals.

Validation Reports 3

Submit, query, and calculate scores and statistics for validation reports.

Dashboard 3

System overview statistics, status distribution, and recent activity tracking.

Setup Steps

1

Configure Your MCP Client

Squid MCP Server uses OAuth2 automatic authentication. Simply add the MCP Server URL — the client will automatically complete the OAuth2 Discovery and authentication flow.

Below are examples for common MCP clients (any MCP protocol-compatible client is supported):

💬

Example: Claude Desktop

Edit the config file:

  • macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
  • Windows: %APPDATA%\Claude\claude_desktop_config.json

Add the following:

{
  "mcpServers": {
    "squid": {
      "url": "https://mcp-server-production-f4cuesozga-de.a.run.app/mcp"
    }
  }
}

Save and restart Claude Desktop. On first connection, the browser will open automatically for OAuth2 login.

⌨️

Example: Claude Code (CLI)

Run the following command in your terminal:

claude mcp add squid \
  https://mcp-server-production-f4cuesozga-de.a.run.app/mcp \
  --transport http

After adding, the OAuth2 flow will trigger automatically the first time you use an MCP tool.

2

Log In & Authorize

The first time you use an MCP tool, the system will guide you through login:

  1. Browser opens automatically
    The MCP client opens your browser and navigates to the Nautilus login page.
  2. Enter your credentials
    Log in with your Nautilus account. If you don't have one, click "Register" on the login page.
  3. See "Authentication Successful"
    After authorization, the browser shows a green checkmark and "Authentication Successful." You can then close the browser.
  4. Return to your client and start using
    Go back to your MCP client — the tools are ready.

After logging in, the system maintains your session automatically. Tokens are refreshed in the background — no re-login required.

3

Verify Connection

After setup, type the following in your AI assistant to test:

Test Command
List all projects

If you see a project list (even empty), the connection is working.

Manual Health Check
curl https://mcp-server-production-f4cuesozga-de.a.run.app/health

Expected response: {"status":"healthy", ...}

Usage Examples

Once connected, use natural language to manage your Squid system through the AI assistant:

Project Management
  • "List all active projects"
  • "Create a project named 'New E-Commerce Platform'"
  • "Change project XXX status to completed"
Requirement Management
  • "List all requirements for project XXX"
  • "Create a requirement ticket titled..."
  • "Change requirement XXX status to in_progress"
Plan Management
  • "Create a development plan for this project"
  • "List all currently active plans"
  • "Change plan XXX status to completed"
Dashboard & Statistics
  • "Show system overview statistics"
  • "Check progress across all projects"
  • "List recent activity records"

Technical Specifications

MCP Protocol Version 2024-11-05
Transport Streamable HTTP
Authentication OAuth2 (Auto Discovery + Bearer Token)
Response Format JSON-RPC 2.0
Content-Type application/json
Accept application/json, text/event-stream
Deployment GCP Cloud Run (asia-east1)
Tool Count 43 MCP Tools

FAQ

Q: Connection refused error

Check if the MCP Server is running:

curl https://mcp-server-production-f4cuesozga-de.a.run.app/health

Expected: {"status":"healthy"}. If no response, please try again later.

Q: 401 Unauthorized error

Check the following in order:

  • Have you completed the login authorization? (Browser should open automatically on first use)
  • Is your Nautilus account approved and activated?
  • If still failing, try removing and re-adding the MCP Server configuration to trigger re-login.

Q: Tools not visible in MCP client

Please verify:

  • Config file format is correct (valid JSON syntax)
  • Client app was fully restarted (not just window closed)
  • Check client log files for error messages

Q: Do I need to log in again periodically?

No. Once logged in, the system maintains your session automatically in the background. Re-login is only required after extended inactivity (over 30 days), at which point the client will automatically open the browser to guide you through login.

Q: Which AI clients are supported?

All MCP protocol-compatible clients are supported, including:

  • Claude Desktop / Claude Code — Anthropic official implementations (desktop / CLI)
  • Third-party MCP clients — any tool supporting MCP Streamable HTTP Transport
  • Custom implementations — anything built to the MCP specification

MCP is an open standard — not tied to any specific AI vendor or platform.

Ready to Get Started?

Connect your AI assistant to Squid now and experience AI-driven development automation.

Learn About Squid