# Anvil MCP server

> An open-source, self-hosted AI coding agent that orchestrates end-to-end development pipelines across multiple repositories using various LLM providers.

- Repository: https://github.com/esanmohammad/Anvil
- Page: https://hostedmcp.org/servers/esanmohammad/anvil
- Transport: —
- Stars: 20
- License: MIT

## Environment variables

| Name | Required | Secret | Description |
|---|---|---|---|
| `OTEL_EXPORTER_OTLP_ENDPOINT` | no | no |  |
| `OTEL_EXPORTER_OTLP_HEADERS` | no | no |  |
| `OTEL_SERVICE_NAME` | no | no |  |

## FAQ

### Can I run Anvil completely offline?

Yes, you can use Ollama for a fully offline experience using your own GPU.

### How does Anvil handle multiple large local models on one GPU?

It uses an exclusive slot serialization feature with a process-local FIFO queue to prevent OOM errors by ensuring only one heavy model is GPU-resident at a time.

### Does Anvil provide cost tracking?

Yes, it includes a live cost ledger that shows per-call, per-stage, and per-run spend in real time.
