# ego MCP server

> Enables AI agents to perform recursive self-improvement by diagnosing performance drops and verifying corrections using OpenTelemetry traces.

- Repository: https://github.com/triageagent-dev/ego
- Page: https://hostedmcp.org/servers/triageagent-dev/ego
- Transport: —
- Stars: 9
- License: Apache-2.0

## Environment variables

| Name | Required | Secret | Description |
|---|---|---|---|
| `EGO_DECISION_API_KEY` | no | no |  |
| `EGO_DECISION_MAX_STATE_BYTES` | no | no |  |
| `EGO_DECISION_MODEL` | no | no |  |
| `EGO_DECISION_URL` | no | no |  |
| `EGO_OTLP_TOKEN` | no | no |  |
| `EGO_SERVER_URL` | no | no |  |
| `GEMINI_API_KEY` | no | no |  |
| `LANGSMITH_TRACING` | no | no |  |
| `OTEL_EXPORTER_OTLP_ENDPOINT` | no | no |  |
| `OTEL_EXPORTER_OTLP_HEADERS` | no | no |  |
| `OTEL_SERVICE_NAME` | no | no |  |

## FAQ

### How does ego facilitate agent self-improvement?

It uses a loop where it samples sessions, scores them, detects regressions, diagnoses the cause, and hands the diagnosis to the agent via MCP for correction and verification.

### Does ego require a specific SDK to instrument my agent?

No, it works with any OpenTelemetry-instrumented agent without requiring an additional SDK or extra runtime.

### What metrics can ego calculate?

It can calculate various metrics including tool trajectory scores, response match scores, hallucinations, safety, and custom Jev decision model metrics like task resolution and grounding.

### Can I use my own decision model for scoring?

Yes, you can point ego to a custom endpoint, such as an Ollama instance, to use as the decision model.