Connecting Your Stack: A Complete Guide to AgentAGI Integrations

Integration Overview

The Integrations page at app.agentagi.dev/integrations is your central hub for connecting external services. AgentAGI supports multiple integration types:

1. Claude Code CLI Integration

The Claude Code CLI integration (507 lines of dedicated adapter code) connects AgentAGI to Claude's terminal-based coding agent. This is one of our most powerful integrations — it lets your AI agents write, review, and deploy code through Claude Code.

How to connect:

1. Go to Integrations page in your dashboard
2. Click “Connect Claude Code”
3. Follow the setup instructions to install the Claude Code CLI on your server
4. AgentAGI handles the rest — agents can now dispatch coding tasks to Claude Code

The integration includes a global execution lock that serializes concurrent agent calls. Since multiple agents share a single Claude account, only one CLI call executes at a time. This prevents conflicts and ensures clean task execution.

2. Ollama Local LLMs

For privacy-sensitive workloads or to reduce API costs, connect Ollama (141 lines of adapter code) to run local LLMs. This lets your agents use models like Llama 3, Mistral, or CodeLlama running on your own hardware.

Setup is straightforward:

1. Install Ollama on your server
2. Pull the models you want to use
3. Enter your Ollama endpoint URL in the Integrations page
4. Assign Ollama models to specific agents based on their needs

3. MCP Protocol (Model Context Protocol)

AgentAGI supports the Model Context Protocol (MCP), the open standard developed by Anthropic for connecting AI agents to external tools and data sources. MCP allows your agents to:

• Query databases — Connect to PostgreSQL, SQLite, or any MCP-compatible database
• Access APIs — Give agents access to external APIs through MCP tool definitions
• Read files — Allow agents to read and write files in your workspace
• Use custom tools — Define custom MCP tools for your specific use cases

To add an MCP integration, navigate to the Integrations page, click “Add MCP Tool,” and enter the tool's endpoint URL. The MCP client service handles the protocol handshake, tool discovery, and secure execution automatically.

4. Plugin Framework

For advanced use cases, build custom plugins using our plugin framework. The plugin system (340-line plugin manager) lets you extend AgentAGI with:

• Custom tools — Add new capabilities that agents can discover and use
• Event hooks — Run custom code when tasks complete, budgets change, or trust scores update
• Agent definitions — Create entirely new agent types with specialized behaviors
• Data sources — Connect proprietary data sources for agent access

See our technical guide for a full walkthrough of plugin development, including the plugin manifest structure and deployment process.

5. Webhooks

Configure webhooks to receive real-time notifications from your AI company. Webhooks are useful for:

• Sending Slack notifications when tasks complete
• Updating your CRM when new leads are generated
• Triggering external workflows when budgets are exhausted
• Logging agent activity to your own monitoring system

// Example: Slack notification webhook
{
  "event": "task.completed",
  "agent": "Lead Researcher",
  "task": "Found 15 new leads for SaaS product",
  "cost": 0.89,
  "company": "My AI Company"
}
→ POST to https://hooks.slack.com/services/...

6. REST API

Every feature in AgentAGI is accessible through the REST API. Generate an API key from your Settings page and start building. The API supports:

• Company and agent management
• Task creation, approval, and tracking
• Budget monitoring and adjustment
• Activity log queries and analytics
• Plugin installation and management

7. OpenAI-Compatible Providers

AgentAGI works with any OpenAI-compatible API endpoint. This means you can connect:

• OpenAI (GPT-4, GPT-4o) — Standard integration
• Anthropic (Claude) — Via API or Claude Code CLI
• Together AI — For open-source model access
• Groq — For high-speed inference
• Custom endpoints — Any OpenAI-compatible API

You can assign different providers to different agents. Use GPT-4 for strategic planning and a cheaper provider for routine tasks — all within the same AI company.

Getting Started with Integrations

Ready to connect your stack? Here's the recommended order:

1. Start with a template — Launch your AI company first with default settings
2. Connect your AI provider — Add your preferred LLM (OpenAI, Claude, Ollama)
3. Add MCP tools — Connect databases and APIs your agents need
4. Set up webhooks — Get notifications in Slack, Discord, or your own system
5. Build custom plugins — Extend AgentAGI for your specific needs