Seeking Advice on MCP Server Design: Stream Handling & Plugin API Best Practices

[caomengxuan666]

Pre-submission Checklist

• I have checked that this question would not be more appropriate as an issue in a specific repository
• I have searched existing discussions and documentation for answers

Question Category

• Protocol Specification
• SDK Usage
• Server Implementation
• General Implementation
• Documentation
• Other

Your Question

I’m developing an MCP server in C++ with a C-compatible API for plugins. Need feedback on streaming/resumption design and protocol compliance.

Current Approach

Session & Stream Management

• Stateful Sessions: Track clients via session_id and stream progress via event_id.
• Example Flow:
  • Client disconnects at event_id=10 (server at 13).
  • Unsent events (11-13) are buffered.
  • On reconnect, client sends its event_id, server replays cached events before resuming.

Plugin API Design

• Non-streaming: Simple request/response.
• Streaming:
  • Plugins register callbacks to emit events.
  • Transport layer handles buffering/retries (plugins unaware of client state).

Questions

1. Protocol Compliance:

   • Is event_id-based resumption standard in MCP? (Java SDKs seem opaque here.)
   • If no formal spec exists, what are best practices for streaming with intermittent connections?

2. Plugin Responsibility:

   • Should plugins manage the session resumption function by themselves?For example.Or just managed by the trasport layer?I doubt that,
     suppose that you are using ChatGpt and you disconnect from it,it still works,and I think write the resume code brings
     more burden for those plugin develops,it seems necessary.

---

C++ Dev Note: I’m prioritizing low-latency and cross-language support (hence C API). Grateful for any insights!