Building an AI Coding Workflow with Claude Code + MCP:......

Claude Code + MCP: Why This Combo Is Worth Your Attention

Claude Code is already powerful on its own — it can understand entire codebases, modify code across files, and execute commands in the terminal. But its “world” is limited to your project folder.

Claude Code lifecycle hooks and MCP integration diagram

The MCP protocol breaks down this wall. Through MCP Servers, Claude Code gains access to your entire development toolchain:

Project Management Systems: Jira, Linear, Notion
Code Hosting Platforms: GitHub, GitLab
Databases: PostgreSQL, SQLite, MongoDB
Monitoring Tools: Sentry, Datadog
Design Tools: Figma
Communication Tools: Slack, Telegram, Discord

You no longer need to manually copy and paste ticket descriptions into chat windows, or take screenshots of database query results to send to AI. Claude Code can directly read and operate these systems.

Step 1: Install Claude Code

If you haven’t installed Claude Code yet, here are several options:

macOS / Linux / WSL (Recommended one-click install):

curl -fsSL https://claude.ai/install.sh | bash

Homebrew (macOS):

brew install --cask claude-code

Windows PowerShell:

irm https://claude.ai/install.ps1 | iex

After installation, run the following in any project directory:

cd your-project
claude

On first run, you’ll be prompted to log in. It supports Claude Pro/Max/Team subscriptions, Anthropic Console API accounts, as well as third-party providers like Amazon Bedrock and Google Vertex AI.

Once logged in, your credentials are saved locally, so you won’t need to log in again.

For more installation details, refer to: Claude Code Official Documentation

Step 2: Understanding MCP’s Three Connection Methods

Claude Code supports three MCP Server transport protocols, each suited for different scenarios:

1. Remote HTTP Server (Recommended)

Ideal for connecting to cloud services, this is currently the most widely supported transport method.

# Connect to Notion MCP
claude mcp add --transport http notion https://mcp.notion.com/mcp

# API connection with authentication
claude mcp add --transport http secure-api https://api.example.com/mcp \
  --header "Authorization: Bearer your-token"

In the .mcp.json configuration file, the type field also accepts streamable-http as an alias for http, so configurations copied from other MCP documentation can be used directly.

2. Remote SSE Server

# Connect to Asana (Note: SSE transport is deprecated, HTTP is preferred)
claude mcp add --transport sse asana https://mcp.asana.com/sse

3. Local stdio Server

Suitable for scenarios requiring direct access to local system resources. Claude Code automatically passes the CLAUDE_PROJECT_DIR environment variable to subprocesses, making it easier for Servers to resolve project-relative paths.

# Connect to Airtable MCP Server
claude mcp add --transport stdio --env AIRTABLE_API_KEY=YOUR_KEY airtable \
  -- npx -y airtable-mcp-server

⚠️ Option order matters: All options (--transport, --env, --scope, --header) must be placed before the Server name. The -- (double dash) is followed by the command and parameters passed to the MCP Server.

Step 3: Hands-On Practice — Building Your Development Workflow

Here’s a complete hands-on scenario: using Claude Code + MCP to automate the workflow from ticket to PR.

Scenario Description

You’re maintaining a web project. The product manager has created a new requirement in Linear, and you need to:

Read the requirement description
Implement the feature in the codebase
Run tests
Commit the code and create a GitHub PR
Notify the team on Slack

Configuring MCP Servers

First, connect GitHub and Linear:

# Connect to GitHub (using personal token)
claude mcp add --transport stdio \
  --env GITHUB_PERSONAL_ACCESS_TOKEN=ghp_xxxx \
  github -- npx -y @anthropic/mcp-server-github

# Connect to Linear
claude mcp add --transport http linear https://mcp.linear.app/mcp \
  --header "Authorization: Bearer lin_api_xxxx"

After connecting, you can confirm the status with:

claude mcp list

Running /mcp in a Claude Code session shows the number of tools and connection status for each Server.

Actual Usage

Now you can complete the entire workflow using natural language:

Read the requirements from Linear ticket WEB-123, implement this feature,
create a feature/web-123 branch on GitHub,
run tests after writing the code, and if they pass, submit a PR and notify the team.

Claude Code will:

Read the ticket details via the Linear MCP Server
Create a Git branch
Understand the requirements and write the code
Run test commands like npm test
Create a Pull Request via the GitHub MCP Server
Send a notification via the Slack MCP Server

Database Query Workflow

If you also need to query the database to assist development:

# Connect to PostgreSQL
claude mcp add --transport stdio \
  --env PG_CONNECTION_STRING=postgresql://user:pass@host:5432/db \
  postgres -- npx -y @modelcontextprotocol/server-postgres

Then you can ask Claude Code:

Query user data for the /api/search endpoint from the past week,
analyze performance bottlenecks, and help me optimize the relevant code.

Step 4: Advanced Tips

Dynamic Tool Updates

Claude Code supports MCP list_changed notifications, allowing MCP Servers to dynamically update the list of available tools without restarting the session. This means you can:

Temporarily enable/disable certain tools during a session
Switch tool sets based on project context
Implement conditional tool exposure

Automatically Generate MCP Servers with Claude

If existing MCP Servers can’t meet your needs, Claude Code itself can help you build one:

# Install the official plugin in a Claude Code session
/plugin install mcp-server-dev@claude-plugins-official

# Reload plugins
/reload-plugins

# Run the build tool
/mcp-server-dev:build-mcp-server

Claude will ask about your use case and then automatically generate scaffold code for either a remote HTTP or local stdio Server.

Project-Level MCP Configuration

MCP configurations can be scoped to specific projects. Create a .mcp.json file in the project root directory, and the configuration will only apply to the current project:

{
  "mcpServers": {
    "github": {
      "command": "npx",
      "args": ["-y", "@anthropic/mcp-server-github"],
      "env": {
        "GITHUB_PERSONAL_ACCESS_TOKEN": "ghp_xxxx"
      }
    }
  }
}

This way, different projects can have completely different MCP Server sets without interfering with each other.

Security Reminders

When connecting to MCP Servers, be sure to pay attention to the following security issues:

Only connect to Servers you trust: Servers that fetch external content may introduce prompt injection risks
Principle of least privilege: Only grant MCP Servers the necessary permissions for their API tokens
Regularly review connected Servers: Use claude mcp list to audit your configuration and remove Servers you no longer use
Sensitive data isolation: For sensitive information like database connection strings, use environment variables instead of hardcoding them

Summary

The combination of Claude Code + MCP essentially solves a core problem: How to transform an AI coding assistant from “can write code” to “can get work done”.

When Claude Code can directly read tickets, query databases, operate Git, and send notifications, it’s no longer a passive Q&A tool but an intelligent agent capable of independently executing complete development workflows.

The barrier to entry for this workflow isn’t high — install Claude Code, connect a few MCP Servers, describe your requirements in natural language, and let AI handle the rest. If you’re still manually copying and pasting ticket content, manually querying databases, and manually creating PRs, this solution is definitely worth trying.

Recommended Reading:

#Claude Code #MCP #AI Coding Workflow #Developer Tools #Anthropic