ai-agent-mcp-server

anirudhatalmale7-a11y/ai-agent-mcp-server

3.2

If you are the rightful owner of ai-agent-mcp-server and would like to certify it and/or have it hosted online, please leave a comment on the right or send an email to dayong@mcphub.com.

The AI Agent MCP Server is a tool designed to facilitate the execution of commands and management of files and repositories on a GCP VM via SSH, using a Claude agent.

Tools
12
Resources
0
Prompts
0

AI Agent MCP Server

Desktop Commander-like tools for Claude agent, executed via SSH on a GCP VM.

Architecture

Claude API (brain)
    ↓
FastAPI MCP Server (Cloud Run)
    ↓
SSH → e2-micro VM (workspace)
    ↓
Git, files, bash, tests

Setup

1. Create GCP VM (Free Tier)

# Create e2-micro in free tier region
gcloud compute instances create claude-workspace \
  --machine-type=e2-micro \
  --zone=us-central1-a \
  --image-family=debian-12 \
  --image-project=debian-cloud \
  --boot-disk-size=10GB

# Get external IP
gcloud compute instances describe claude-workspace \
  --zone=us-central1-a \
  --format='get(networkInterfaces[0].accessConfigs[0].natIP)'

2. Setup VM

# SSH into VM
gcloud compute ssh claude-workspace --zone=us-central1-a

# Run setup script
curl -sSL https://raw.githubusercontent.com/YOUR_REPO/vm-setup.sh | bash

3. Configure SSH Keys

# Generate key pair (local machine)
ssh-keygen -t rsa -b 4096 -f mcp_server_key -N ''

# Copy public key to VM
gcloud compute ssh claude-workspace --zone=us-central1-a --command="
  echo '$(cat mcp_server_key.pub)' >> /home/claude-agent/.ssh/authorized_keys
  chmod 600 /home/claude-agent/.ssh/authorized_keys
  chown claude-agent:claude-agent /home/claude-agent/.ssh/authorized_keys
"

4. Store Secrets in GCP Secret Manager

# VM IP
echo -n "YOUR_VM_IP" | gcloud secrets create VM_HOST --data-file=-

# SSH private key
gcloud secrets create VM_SSH_KEY --data-file=mcp_server_key

# MCP API Key (optional)
echo -n "your-secure-api-key" | gcloud secrets create MCP_API_KEY --data-file=-

5. Deploy MCP Server

# Trigger Cloud Build
gcloud builds submit --config=cloudbuild.yaml

Usage

Direct API Calls

MCP_URL="https://ai-agent-mcp-server-xxx.run.app"

# Execute bash
curl -X POST "$MCP_URL/tools/bash" \
  -H "Content-Type: application/json" \
  -d '{"command": "ls -la"}'

# Read file
curl -X POST "$MCP_URL/tools/read_file" \
  -H "Content-Type: application/json" \
  -d '{"path": "README.md"}'

# Clone repo
curl -X POST "$MCP_URL/tools/git/clone" \
  -H "Content-Type: application/json" \
  -d '{"repo_url": "https://github.com/user/repo.git"}'

Python Agent

export MCP_SERVER_URL="https://ai-agent-mcp-server-xxx.run.app"
export ANTHROPIC_API_KEY="your-key"

python agent.py "Clone the repo, fix the bug in main.py, test it, and push"

Tools Available

ToolDescription
bashExecute shell commands
read_fileRead file contents
write_fileCreate/overwrite file
edit_fileSurgical string replacement
delete_fileDelete file
list_directoryList directory
create_directoryCreate directory
git_cloneClone repository
git_pullPull changes
git_commitCommit all changes
git_pushPush to remote
git_statusGet status

Security

  • Path validation: All paths must be within /home/claude-agent/workspaces
  • Command filtering: Dangerous commands blocked (rm -rf /, fork bombs, etc.)
  • API key auth: Optional X-API-Key header
  • SSH key auth: Only MCP server can connect to VM

Cost

  • e2-micro VM: $0 (free tier in us-central1)
  • Cloud Run MCP Server: ~$0-1/month (min-instances=0, only charges during use)