Resumable WebSocket streams

Add MCP server to your AI tool

Allow AI tools and LLMs to interact with the API documentation portal through MCP.

MCP server URL

https://bump.sh/tie-tech/hub/b2b-api/doc/broker-overview/mcp

Standard setup for AI tools providing an mcp.json file

mcp.json
{
  "Overview MCP server": {
    "url": "https://bump.sh/tie-tech/hub/b2b-api/doc/broker-overview/mcp"
  }
}

Close

The /orders, /accounting, and /events streams use the same resume protocol.

  • Data-frame seq values are strictly increasing in delivery order but are not guaranteed to be contiguous. Filters and retention can produce valid gaps. A gap alone is not evidence of data loss and must not trigger a reconnect.
  • Drop a data frame when seq <= last_applied; otherwise apply it atomically with persisting the new last_applied.
  • heartbeat and backlog_end do not advance the data sequence; they repeat the latest relevant seq.
  • Reconnect with since=last_applied. The server may replay that sequence inclusively, then sends backlog_end and continues live delivery. The duplicate check makes this safe.
  • Replay is limited to messages from the preceding one hour. Reconnect promptly and do not rely on older history being available.
  • resume_too_old followed by close code 4009 means the requested position is outside the supported replay window or is otherwise unavailable.
  • Close code 4008 means the client is reading too slowly. Keep socket reads independent from slower business processing and enforce bounded local queues.
  • Use heartbeat interval_sec for liveness detection; reconnect if no frame arrives within the client-defined timeout.

Available URL filters differ by stream:

  • /orders: repeatable market, event_id, and account_id.
  • /accounting: repeatable market and account_id.
  • /events: repeatable event_id, market_id, and lang.

Changing filters changes which sequence numbers are delivered and invalidates assumptions tied to the previous subscription. Persist a resume position together with its filter set.