Kick off a long-running deep-research run on a fresh DeerFlow thread. Returns immediately with the thread/run ids and a web URL; poll deerflow_run_status until it reaches a terminal status, then call deerflow_get_report to read the findings. Deep research takes minutes to ~45 minutes and never blocks this call.

Send a message to a DeerFlow thread and start a run. Omit thread_id to create a new thread, or pass an existing thread_id to continue a conversation. Returns immediately with the thread/run ids and a web URL; poll deerflow_run_status, then deerflow_get_report.

Check the status of a DeerFlow run. Optionally wait up to wait_seconds (capped at 30s) for it to reach a terminal status before returning, to reduce polling round-trips. Terminal statuses: success, error, timeout, interrupted. Also returns live counters (llm_call_count, message_count, total_tokens) and elapsed/last-update times: the counters advance while the run is working, so if they stop moving for several minutes the run may be stalled — use deerflow_run_progress or deerflow_wait_activity for event-level detail.

Get live progress for a DeerFlow run: status + live counters (llm_call_count, message_count, total_tokens), recent activity (the latest events as one-line summaries, e.g. tool calls and their results), the plan-mode todo checklist, and stall/quiet detection (stalled: true when no activity for longer than the stall threshold; quiet: true for a softer 'between steps' signal; next_step: a human hint pointing at the web UI and deerflow_cancel_run). Pass since_seq (the last_event_seq from a previous response) to return only new events. Requires session or internal-token auth: the event stream and thread state are not reachable with a Personal Access Token.

Block server-side until the run produces new activity, reaches a terminal status, or the timeout elapses — one call replaces many status polls. The server joins the run's live event stream (falling back to polling when the stream is unavailable), so it returns the moment new activity appears rather than on a fixed tick. Returns reason ('terminal' | 'activity' | 'timeout'), waited_seconds, timeout_seconds, the new activity since since_seq (one-line summaries), the plan-mode todo checklist, quiet/stall detection, and a next_step hint. Pass the returned last_event_seq as since_seq on the next call to continue from where you left off. While waiting it emits MCP progress notifications (elapsed/timeout) when the client supplies a progress token. Cancelling the request from the client aborts the wait and returns stop_reason 'cancelled_by_client'. Requires session or internal-token auth.

Fetch the synthesized report for a DeerFlow thread: the most recent assistant message, its title, and any produced artifact file paths. Resolves the report text through a fallback chain (run messages → thread state → summary) and, when no assistant message exists, auto-inlines the first text artifact (≤256 KB) as the report. Also reports the run's terminal status and where the text came from (report_source). Call after a run reaches a terminal status. Optionally pass run_id to scope the report to a specific run.

List recent DeerFlow threads (id, title, status, timestamps). Use the returned thread_id with deerflow_chat to continue a conversation or deerflow_get_report to read a finished one.

Cancel an in-flight DeerFlow run (interrupts it). Returns the accepted status.

List the artifact file paths produced by a DeerFlow thread (e.g. reports, generated files). Pass a path to deerflow_get_artifact to read its content.

Fetch a single artifact file from a DeerFlow thread. Text-like files (markdown, json, csv, plain text) are returned inline as content; binary files return a URL reference instead. Note: with a Personal Access Token this endpoint is not in the PAT route allowlist — an internal token is required.

List the models configured on the DeerFlow instance (name, display name, and capability flags). Use a returned name for the model argument of deerflow_research / deerflow_chat. Note: with a Personal Access Token this endpoint is not in the PAT route allowlist — an internal token is required.