mq-bridge-app

Open this server's agent inbox so other agents on this machine can send it messages, and tail it for the rest of the session. Only exists when the server was started with `--agent-bus`, and off until called — nothing can reach this agent before that. `name` is what peers address it by. Messages are then collected with `route_messages` on route `agent-inbox`; reads are destructive, and only the last `capture_last` (default 200) are held — the durable inbox is drained into that buffer, so mail beyond it is discarded, not queued. By default the inbox is a local directory under `$MQB_AGENTS_DIR` (default `~/.mqb-agents`), which any process able to write a file can post to; pass `input` to receive over a broker instead. Calling it twice is an error — one inbox per server.

Send a message to another agent's inbox on this machine. Always available: sending needs no opt-in, because it only writes into a mailbox the recipient chose to open with `agent_listen`. `to` is the peer's name — `server_info` lists the peers that have an inbox and which are currently listening. Delivery is a single atomically written file, so a peer that is not running collects the message when it next listens. Pass `output` instead of `to` to reach an agent over a broker rather than this filesystem. A sender that has not called `agent_listen` is delivered as `from: "unnamed"` and cannot be replied to.

Generate a copyable, headless `mqb` command for a route. Accepts the same route JSON and tuning overrides as `start_route`, but does not start anything. The command embeds canonical JSON with POSIX shell quoting so fields that have no compact copy-URI representation are preserved. Credential values are replaced by environment variable placeholders and the required variable names are returned separately.

List the routes started by this server, with their live connection health, total messages moved, current message rate, and achieved average rate.

Publish one message or a batch of messages to any endpoint. The target is supplied inline as endpoint JSON keyed by type (e.g. {"kafka": {"url": "...", "topic": "..."}}); this is independent of routes. Provide either `message` or `messages`; each message has a `payload` and optional `metadata` headers (e.g. `kind` for the message type, `correlation_id`). Returns the send outcome (an acknowledgement, or the endpoint's reply for request/reply endpoints).

Return the most recent messages captured on a route, newest last. Only routes started with `capture_last` capture anything. Reads are destructive: each message is returned once, so repeated calls yield only what arrived since the last call.

Report the live status of one route (by `name`) or all routes: connection health, whether it has finished, the total number of messages it has moved, its current rate (`messages_per_second`), and the rate it achieved overall (`average_messages_per_second` over `elapsed_s`). Quote the average for a finished job: the current rate decays to ~0 once a route stops moving data.

Report this MCP server's build identity: crate `version`, `git_hash` (with a `-dirty` suffix if the working tree had uncommitted changes at build time), build `profile` (debug/release), and `build_time`. Use it to confirm which binary is actually running before trusting throughput numbers.

Create and run a route that moves messages from its `input` (source) endpoint to its `output` (sink) endpoint, each keyed by connector type. Supports route options like `concurrency`, `batch_size`, and `exit_on_empty` (drain the source then exit; otherwise the route runs continuously until stopped). Either endpoint may carry a `middlewares` array — `retry`, `dlq`, `deduplication`, `limiter`, `transform`, `compression`, `encryption`, `pack`/`unpack`, `buffer`, `delay`, `timeout`, `weak_join`, `cookie_jar`, `metrics` — e.g. `{"sqlx": {...}, "middlewares": [{"retry": {"max_attempts": 3}}]}`. `concurrency`/`batch_size` may be given inside `route` or as top-level arguments — the top-level argument wins, and an unset value becomes the app default (4 / 1024). Returns the route name for use with wait_route / route_status / stop_route, plus the tuning that took effect.

Stop a running route by `name`. Returns the total messages it moved plus the rate it achieved (`average_messages_per_second` over `elapsed_s`).

Block until a route finishes, then report how it ended. Use this instead of polling `route_status` for a job started with `exit_on_empty`: it costs one tool call regardless of how long the job runs. Returns `finished` (false if `timeout_ms` elapsed first — the route is left running), the `outcome`, the total messages moved and the rate achieved. An `outcome` of `failed` carries the cause in `error`. Default timeout 60000 ms, maximum 18000000 ms.