Autonomous Agents & MCP Integration
Exovon is engineered with native Model Context Protocol (MCP) tooling. AI agents (Antigravity, Cursor, Claude Desktop, autonomous swarms) can deploy code, inspect build compiler errors, run database migrations, and execute zero-downtime rollbacks autonomously.
Official MCP Server
Attach @exovon/mcp-server directly to Antigravity, Cursor, or Claude Desktop for structured tool calling.
Self-Healing Feedback Loops
Agents read build failure stack traces, adjust code or configurations, and re-trigger deployments until health checks succeed.
Runaway Guardrails
Rate-limiting quotas, build circuit breakers, and instant token revocation prevent accidental infinite loops from draining budgets.
1. Connecting via Model Context Protocol (MCP)
The Model Context Protocol allows AI assistants to invoke platform operations. Add the Exovon MCP server to your agent configuration file:
{
"mcpServers": {
"exovon": {
"command": "npx",
"args": ["-y", "@exovon/mcp-server@latest"],
"env": {
"EXOVON_API_KEY": "exo_agent_...",
"EXOVON_DEFAULT_PROJECT": "proj_my_app",
"EXOVON_REGION": "asia-south1"
}
}
}
}Exposed MCP Tool Endpoints
| Tool Name | Arguments | Functionality |
|---|---|---|
| exovon_deploy | directory, framework, envVars | Packages code, dispatches build pipeline, returns deploymentId |
| exovon_get_logs | deploymentId, tailLines | Streams recent compiler, container stdout, and stderr logs |
| exovon_database_query | query, params | Executes read-only or migration SQL statements in ExoStore |
| exovon_rollback | targetDeploymentId, reason | Points Edge Router to a healthy prior deployment in <3s |
| exovon_status | projectId | Reports active CPU, memory, bandwidth, and TLS certificate status |
2. Autonomous Security & Safety Guardrails
Enforcing strict policies when AI agents possess deployment write access:
Scoped Agent Tokens
Create scoped agent keys prefixed with exo_agent_.... Restrict access strictly to designated projects and disallow destructive commands like database deletion.
Build Circuit Breakers
Autonomous API keys are bounded to a maximum of 10 builds per hour by default. If an agent enters an infinite loop, the control plane pauses the key and notifies the account owner.
gVisor Microkernel Virtualization
Code generated by agents executes inside isolated gVisor user-space microkernels. All host system calls are intercepted, preventing hallucinated or hostile scripts from escaping containers.
Instant Global Revocation
Revoking a key in the dashboard purges agent authentication across all global Anycast nodes in <500ms, immediately terminating any active deployment or operation.
3. Recommended Directives for .cursorrules / System Prompts
Include these directives in your project instruction files for optimal platform compatibility:
# Exovon Agent Directives
# 1. For Next.js, always verify output: "standalone" is configured in next.config.
# 2. Ensure Node.js HTTP servers bind to process.env.PORT || 8080 on 0.0.0.0.
# 3. For database queries, use @neondatabase/serverless over HTTP to eliminate socket leaks.
# 4. If a build fails, inspect exovon_get_logs output before applying source code modifications.
# 5. Never hardcode secrets in source files; manage credentials via Exovon env commands.Frequently Asked Questions
Can an AI agent create new PostgreSQL clusters?
Only if the API key has been explicitly granted database provisioning scopes. By default, agent tokens are restricted to querying or running migrations on pre-existing databases.
What happens if an agent deploys code that crashes in production?
Exovon canary traffic testing verifies new revisions against health checks before cutting over traffic. If 5xx error thresholds are triggered, the router triggers an instant rollback to the previous deployment.