The MCP gateway space is crowded, and most projects are good at different things. This is an honest map: UMB's edge is context efficiency and a single source-available binary — not governance, where other gateways lead. Facts reflect each project's own public positioning.
| GATEWAY | TOKEN REDUCTION | LICENSE | SINGLE BINARY | PRIMARY FOCUS |
|---|---|---|---|---|
| Universal MCP Bridge | ~99.1% (measured, 138,417-token baseline) | PolyForm Noncommercial 1.0.0 (source-available) | Yes — one Rust binary | Context efficiency |
| AIRIS MCP Gateway | ~97% (third-party listings; no first-party benchmark published) | See project | No (gateway service) | Context efficiency |
| mcpproxy-go | BM25 tool filtering (no single % published) | Open source | Yes — Go binary | Tool filtering / search |
| MetaMCP | Not the headline metric | Open source | No (self-hosted app) | Consolidation / governance |
| Smithery | Not the headline metric | Hosted platform | No (hosted registry/runtime) | Registry / hosting |
| mcp-proxy | Not a goal (transport bridge) | Open source | Varies | stdio ↔ HTTP transport |
FIGURES REFLECT EACH PROJECT'S PUBLISHED POSITIONING · UMB FIGURE IS MEASURED
If your problem is auth, policy, multi-tenant access control, or a hosted registry, a governance-focused gateway is the right tool — and UMB doesn't try to outdo them there.
If your problem is that connecting more MCP servers keeps eating your agent's context window, UMB is built for exactly that: a ~99.1% measured cut in static tool context, a single source-available Rust binary, and no feature gates — every capability ships in every copy.