GarmX

Project Status: early stage release. The design is settled; implementation is in progress; can use project for early build - it is usable and available

GarmX is the local-first observability & audit plane for MCP connections — see, audit, and trust what your AI coding agents and agentic apps actually do across every tool. OTel export. All in a single lean Go binary.

To any AI client, GarmX presents itself as one MCP server — it merges every registered upstream's tools into a single catalog, routes each MCP call to the right server, and records every transaction along the way.

Key strengths — observability & audit

Every AI client -> MCP transaction is captured and logged for auditing, can be vieved in a lean web UI, and exported over OpenTelemetry to Grafana / Prometheus / Loki / any platform which supports OTLP.

This is cross-server, cross-agent view.

Also:

• Consolidate — garmx import sweeps servers scattered across your client configs into one catalog; you then point every AI client at just garmx.
• Curate — per-agent profiles can expose the right number of tools you choose, not all of them at once: better tool-selection, lower token cost.
• Safe by default — secrets and other sensitive data you choose are redacted before they reach the audit store; garmx daemon only binds to 127.0.0.1

Quick start

make build          # compile to bin/garmx
./bin/garmx -h      # usage
make check          # full validation gate (run after every change)
make run            # build + run the daemon

The garmx daemon binds 127.0.0.1:9735 by default — it holds every upstream's credentials and must not be exposed publicly.

Development

For GarmX developers; Makefile includes multiple methods to validate, test and build the project.

make check (fmt → lint → vet → test → build) is the quality gate; validate project with it after every change.

Dev tools (gofumpt, golangci-lint, templ) are invoked via go run with the pinned versions, so no global installs are required.

Standards and conventions are outlined in the AI coding agent standards: AGENTS.md.

Documentation

While project is eveolving, concepts, design discussions and implementation plan are being captured in the living docs:

• docs/architecture.md — system design, aggregation model, package layout, schemas.
• docs/implementation.md — phased build and implementation plan.
• docs/discovery.md — open research, decisions, and scope management.

Contributing

Contributions are welcome — see CONTRIBUTING.md.

License

GarmX is licensed under the Apache License 2.0 — see LICENSE.