FAQ
Quick answers to the questions that come up before the deeper guides do. For the full positioning story, see Why MCP REST Bridge →.
Do I need Kubernetes or a separate database?
No. The bridge is a single Bun process with bun:sqlite — one binary, one file. It runs fine inside a container orchestrator if you already have one, but nothing about it requires Kubernetes, Postgres, or any external service. See Deployment →.
Does it work with Claude Desktop, Cursor, or other MCP clients?
Yes — any client that speaks the Model Context Protocol. The bridge negotiates the protocol version via the official TypeScript SDK, which supports 2024-10-07 through 2025-11-25 (defaulting to 2025-03-26 when a client sends none), and is tested against Claude Desktop, Cursor, and custom agents. The gateway connect CLI command generates ready-to-paste config for Claude Desktop, Cursor, Windsurf, and Continue specifically. See Connecting MCP clients →.
Is there a hosted/managed version?
No — this is self-hosted, MIT-licensed open source with no managed SaaS offering or SLA. You run it; you own the data.
How is this different from an OpenAPI-to-MCP CLI tool?
Those tools are great for a one-shot spec-to-tools translation, but they don't manage a running fleet — no admin UI, no per-tool policy, no audit log, no way to aggregate multiple backends behind one governed endpoint. See the comparison table in Why MCP REST Bridge → for the fuller picture.
Can it aggregate multiple MCP servers, not just REST APIs?
Yes — register an existing MCP server as an upstream (kind: "mcp") and it's re-exposed through the same guard stack as a REST-derived backend: same guardrails, same RBAC, same audit log. See Registering backends →. To expose tools from several backends (REST and/or MCP) through one MCP endpoint, curate a bundle →.
What happens if a backend goes down?
Health checks auto-evict unreachable backends, and per-client circuit breakers fail calls fast once a backend is clearly unhealthy instead of piling up timeouts. If you've configured a canary/failover secondary, traffic can route there automatically. See Guardrails & resilience →.
Can I manage config as code instead of clicking through the UI?
Yes — the bundled gateway CLI (bun run cli) treats configuration as a reviewable YAML file: pull the live config, edit it, plan to preview drift, apply to push it. Useful for CI/CD and GitOps-style review. See CLI →.
How do upgrades work?
Schema migrations run automatically on startup and are forward-only — there's no automated rollback, so back up your SQLite database before upgrading to a new version. See Deployment → Upgrading.
Can I try it without installing anything?
Yes — the live demo runs the real admin UI against mock data, right in your browser. No install, no signup.
The demo is fully localized to Spanish as well — flip the locale switcher in Account → Preferences → Language and every translatable fixture string (tool descriptions, bundle summaries, API key labels, alert names, team / policy / composite / snapshot labels, catalog descriptions) updates in place without losing your tab state.
What license is this?
MIT. See LICENSE.
Something's not working — where do I look?
See Troubleshooting → for the common issues (most are deliberate safety behavior rather than bugs), or open an issue with the request_id from the error response.
Next: Getting started → · Contributing →