Amp

Amp is a frontier coding agent that runs in your terminal and editor, combining state-of-the-art AI models with native Model Context Protocol (MCP) support. Built by Sourcegraph, Amp provides a multi-model approach using Claude, GPT-5, and fast models for what each is best at, with seamless MCP integration for extending AI capabilities.

Why Use Amp with MCP?

Amp offers a streamlined MCP experience designed for developers:

Multi-Transport Support: Connect to stdio, SSE, and HTTP-based MCP servers
CLI and GUI Configuration: Add servers via amp mcp add command or settings.json
OAuth Integration: Built-in OAuth flow for authenticated MCP servers
Permission System: Fine-grained control over which tools can execute
Thread-Based Workflow: Save, share, and resume AI conversations with full tool context
Editor Integration: Seamless MCP access from VS Code, Cursor, JetBrains, and Neovim

Key Features

Multi-Model Intelligence

Amp uses the right model for each task:

Smart Mode: Unconstrained state-of-the-art model use (Claude Opus 4.5, GPT-5)
Rush Mode: Faster, cheaper models for well-defined tasks
Free Mode: Free tier using fast basic models

Native MCP Support

Connect to any MCP server with flexible configuration:

{
  "amp.mcpServers": {
    "playwright": {
      "command": "npx",
      "args": ["-y", "@playwright/mcp@latest", "--headless"]
    },
    "custom-api": {
      "url": "https://api.example.com/mcp",
      "headers": {
        "Authorization": "Bearer ${API_TOKEN}"
      }
    }
  }
}

Thread-Based Conversations

Persistent Threads: All conversations saved and resumable
Thread Sharing: Share threads with team members via URLs
Context Preservation: MCP tool usage preserved across sessions
Version Control for AI: Track changes and decisions over time

AGENTS.md Guidance

Amp automatically reads AGENTS.md files for:

Codebase structure and conventions
Build and test commands
Tool usage guidelines
Project-specific instructions

Permission System

Fine-grained control over tool execution:

{
  "amp.permissions": [
    {"tool": "mcp__playwright_*", "action": "allow"},
    {"tool": "mcp__*", "action": "ask"},
    {"tool": "Bash", "matches": {"cmd": "*git push*"}, "action": "ask"}
  ]
}

Installation

CLI Installation (Recommended)

# macOS/Linux/WSL
curl -fsSL https://ampcode.com/install.sh | bash

# Or via npm
npm install -g @sourcegraph/amp

Editor Extensions

VS Code / Cursor:

Install sourcegraph.amp from VS Code Marketplace or Open VSX Registry

JetBrains (IntelliJ, WebStorm, GoLand, etc.):

amp --jetbrains

Neovim:

Install Amp CLI and amp.nvim plugin

MCP Configuration

Method 1: CLI Commands

# Add HTTP-based server with headers
amp mcp add sourcegraph \
  --header "Authorization=token your-token" \
  https://sourcegraph.example.com/.api/mcp/v1

# Add SSE-based server
amp mcp add linear https://mcp.linear.app/sse

# OAuth authentication
amp mcp oauth login my-server \
  --server-url https://example.com/.api/mcp/v1 \
  --client-id your-client-id \
  --client-secret your-client-secret \
  --scopes "openid,profile,email"

Method 2: Settings Configuration

Add to VS Code/Cursor settings.json:

{
  "amp.mcpServers": {
    "filesystem": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-filesystem", "/path/to/files"]
    },
    "github": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-github"],
      "env": {
        "GITHUB_TOKEN": "${GITHUB_TOKEN}"
      }
    },
    "remote-service": {
      "url": "https://mcp.service.com/mcp",
      "headers": {
        "Authorization": "Bearer ${SERVICE_TOKEN}"
      }
    }
  }
}

MCP Permission Controls

Amp provides granular permission controls for MCP tools:

{
  "amp.mcpPermissions": [
    {
      "matches": {"command": "npx", "args": "* @playwright/mcp@*"},
      "action": "allow"
    },
    {
      "matches": {"url": "https://mcp.trusted.com/mcp"},
      "action": "allow"
    },
    {
      "matches": {"command": "*"},
      "action": "reject"
    }
  ]
}

Using Amp

Interactive Mode

# Start interactive session
amp

# With specific MCP config
amp --mcp-config '{"server": {"command": "npx", "args": ["-y", "mcp-server"]}}'

Execute Mode

# Single command execution
amp -x "what files in this folder are markdown files?"

# Piped input
echo "analyze this code" | amp -x

# Allow all tools without prompting
amp --dangerously-allow-all -x "Run sed to replace 2024 with 2025"

Thread Management

# Continue previous thread
amp threads continue --execute "now add 8 to that"

# Stream JSON output for automation
amp --execute "analyze code" --stream-json

Agent Modes

| Mode | Description | Cost | |------|-------------|------| | Smart | State-of-the-art models (Claude Opus 4.5), maximum capability | Paid credits | | Rush | Faster, cheaper models for well-defined tasks | Reduced credits | | Free | Basic models, no charge | Free |

Switch modes via command palette (Ctrl+O → mode) or editor UI.

Prompting Best Practices

Be explicit: Instead of "can you do X?", say "do X"
Keep focused: One task per thread, avoid mixing unrelated work
Provide context: Mention files to look at, commands to run
Use AGENTS.md: Document build/test commands for consistent results
Abandon noisy threads: Start fresh if too many failed attempts clutter context

Use Cases

Development Workflow

Integrate GitHub, Linear, filesystem access, and code analysis tools for comprehensive development assistance.

Code Review & Analysis

Use MCP servers for static analysis, security scanning, and code quality checks.

Research & Documentation

Connect web search, documentation, and knowledge base tools for research tasks.

Automation

Pipe commands and scripts through Amp with MCP tools for automated workflows.

Platform Support

| Platform | CLI | VS Code/Cursor | JetBrains | Neovim | |----------|-----|----------------|-----------|--------| | macOS | ✅ | ✅ | ✅ | ✅ | | Linux | ✅ | ✅ | ✅ | ✅ | | Windows (WSL) | ✅ | ✅ | ✅ | ✅ |

Pricing

Free Mode: No charge, uses fast basic models
Smart/Rush Mode: Credits-based pricing, pay for usage
Sign up: ampcode.com/install

Amp

Amp