Skip to content

Choosing an MCP Connection Model ​

Choose based on who owns configuration, routing, and operations. Server count alone does not determine the right approach.

Compare responsibilities ​

ApproachUseful whenResponsibilities and tradeoffs
Direct MCP connectionsYour client already manages the backends you needConfigure and authenticate each connection using the client's supported features; no extra proxy process
1MCPYou want a shared inventory, filtering, CLI discovery, or project-aware backendsOperate the runtime and its backends, secure access, and verify client compatibility
Custom proxyYou need behavior existing implementations cannot provideImplement, test, secure, and maintain the required protocol and routing behavior
Managed proxyYou want a provider to operate part of the infrastructureVerify the specific provider's features, data handling, pricing, support, and service commitments

This is a comparison of architectural choices, not a benchmark or vendor survey. It does not establish cost savings, setup times, uptime, or certification.

What 1MCP provides ​

The proxy adds a process and a network or transport hop. Local STDIO backends still run as child processes; aggregation does not eliminate their memory or CPU needs. 1MCP does not provide connection pooling, distributed clustering, or an uptime SLA. Compatibility must be checked against the clients, protocol features, and backends you actually use.

Evaluate with one backend ​

  1. Follow the Quick start with one trusted backend.
  2. Confirm that the intended client can discover and invoke its capabilities.
  3. Check authentication, filtering, failure handling, and resource use for your workload.
  4. Add more backends only after the first path works.

For existing desktop configurations, see App consolidation. Retain a rollback path and verify the replacement connection before removing the previous setup.

Released under the Apache 2.0 License.