DocumentationMumbai
Autonomous Cloud
MCP CompliantAgent-Native

Autonomous Agents & MCP Integration

Exovon is the first cloud platform designed with a native Model Context Protocol (MCP) interface. Whether you use Antigravity, Cursor, Claude Desktop, or custom autonomous multi-agent swarms, agents can deploy code, inspect build errors, run SQL migrations, and execute zero-downtime rollbacks autonomously.

Official MCP Server

Plug @exovon/mcp-server directly into Cursor, Claude Desktop, or AGY for standard tool calling.

Self-Healing Loops

Agents read build failure stack traces, adjust project configs, and redeploy automatically until production passes health checks.

Runaway Guardrails

Hardware spend limits, burst rate-limiting, and token kill-switches prevent runaway loops from exhausting cloud budgets.

1. Connecting Exovon via Model Context Protocol (MCP)

The Model Context Protocol allows frontier AI models to call structured functions against external services. Add the Exovon MCP server to your agent configuration:

// ~/.config/Claude/claude_desktop_config.json OR .cursor/mcp.json
{
  "mcpServers": {
    "exovon": {
      "command": "npx",
      "args": ["-y", "@exovon/mcp-server@latest"],
      "env": {
        "EXOVON_API_KEY": "exo_agent_sec_9918237498a...",
        "EXOVON_DEFAULT_PROJECT": "proj_01J8ABC123",
        "EXOVON_REGION": "asia-south1"
      }
    }
  }
}

Tools Exposed to the Agent

MCP Tool NameParametersDescription
exovon_deploydirectory, framework, envVarsBundles code, triggers build pipeline, returns deploymentId
exovon_get_logsdeploymentId, tailLinesFetches recent compiler/container stdout & stderr logs
exovon_database_queryquery, paramsRuns read-only or migration SQL statements in ExoStore
exovon_rollbacktargetDeploymentId, reasonSwitches Edge Router to a healthy previous build in < 3s
exovon_statusprojectIdReturns CPU, memory, edge bandwidth, and SSL status

2. The Self-Healing Deployment Loop

Because Exovon exposes granular build phases via streaming APIs, AI agents can execute automated repair loops without human supervision:

1

Agent Bundles & Deploys

The agent finishes generating features, calls exovon_deploy, and receives a unique build ID.

2

Build Failure Detection

If the Next.js compiler errors (e.g. Module not found: @lucide/react or missing dynamic export), the build fails with exit code 1.

3

Log Extraction & AST Repair

The agent calls exovon_get_logs, extracts the exact line number, installs the missing dependency or fixes the invalid syntax in your workspace.

4

Autonomous Verification

The agent triggers a patch build. Once the Edge Router returns 200 OK, the agent alerts the developer that the feature is live.

3. Security & Runaway Guardrails

Granting autonomous systems write access to cloud infrastructure requires strict, non-negotiable security boundaries. Exovon enforces the following hardware and software policies:

Scoped Agent Tokens

Never share your account owner key. Issue dedicated agent keys prefixed with exo_agent_.... You can constrain an agent to a single project and forbid destructive actions like deleting databases.

Build Rate Limits & Cost Circuit Breakers

Exovon caps autonomous API keys to a maximum of 10 builds per hour by default. If an agent enters an infinite retry loop, the control plane suspends the token and emails the account owner.

gVisor Microkernel Isolation

All applications compiled by agents execute inside gVisor microkernels. The runtime intercepts all host kernel calls, ensuring malicious or hallucinated code cannot escape or compromise other tenant workloads.

Instant Global Revocation

If an agent behaves erratically, clicking "Revoke" in the web console purges the credentials across all worldwide Anycast nodes in under 500ms, immediately killing any active build or deployment.

4. Recommended Agent Instructions (.cursorrules / System Prompt)

Add these directives to your repository's .cursorrules or your custom agent system prompt for optimal Exovon compatibility:

# Exovon Agent Rules
1. When deploying Next.js, always ensure output: "standalone" is in next.config.mjs.
2. Always bind Node.js HTTP servers to process.env.PORT || 8080 on host "0.0.0.0".
3. For database access, use @neondatabase/serverless over HTTP instead of pg socket pools.
4. When a deployment fails, use exovon_get_logs to analyze the last 50 lines before guessing.
5. Never hardcode credentials into source files; use exovon environment variable commands.

Frequently Asked Questions

Can an autonomous agent create new PostgreSQL databases?

Only if the API key has been explicitly granted the 'database:provision' scope. By default, agent keys are restricted to read-only queries or migrations on pre-existing databases.

What happens if an agent pushes code that crashes in production?

Exovon's canary traffic gate tests new builds against active HTTP health probes before cutting over 100% of user traffic. If 5xx errors exceed 1%, the Edge Router automatically triggers an instant rollback to the previous deployment.

Was this documentation page helpful?