MCP tool reference

Connect to https://mcp.jellypod.com/mcp and call tools/list to read the current input and output schemas. The tool set changes in place on this endpoint. Refresh cached tool definitions before updating an existing integration. REST endpoint names and request bodies remain unchanged.

Tools

Grouped tools use explicit kind and id arguments. The resource kinds are podcast, episode, host, source, video, short, and voiceover; each tool's schema restricts which kinds it supports. Do not infer that a kind supported by a read tool is also supported by a destructive tool.

list_resources takes kind, with optional cursor, limit, and podcast_id for episode lists. Episode filters use episode_status; Source filters use source_status and source_type. Results return rows, hasMore, and nextCursor; pass nextCursor as cursor for the next page.

get_resource returns an envelope with kind, id, and resource, plus progress, delivery, transcript, or studio_url when applicable. Keep resource and progress fields in their returned shape rather than assuming all products share one status enum. For polling, read progress.poll_after_seconds. Product delivery uses url; episode delivery uses audio_url and video_url.

For episode transcripts, set transcript_format to json, srt, or vtt. transcript_limit bounds words for JSON or characters for captions; continue with transcript.next_offset as transcript_offset until it is null. Omit transcript fields during ordinary progress polling.

delete_resource supports only podcasts, episodes, hosts, and Sources. Use force: true for published or scheduled content only after explicit user confirmation. Podcast deletion cascades to its episodes. retry_generation supports only Videos, Shorts, and Voiceovers.

save_podcast and save_host use mode: "create" or "update", with id required for an update. Their other input fields are at the top level. set_episode_publication uses id and mode: "publish", "schedule", or "unpublish"; scheduling also requires scheduled_time.

Use generate_episode with mode: "brief" for generated content or mode: "script" for supplied narration. Read the mode-specific schema before calling it. For complete sequences, see the agent guide.

request_source_upload uses mode: "direct" by default, requiring file name, MIME type, and size. mode: "picker" needs no file metadata and returns a handoff with studio_url; its source and upload fields are absent until a file is uploaded. Preparation is automatic in both modes. Pass the completed upload’s Source ID to generation; it waits for Source processing as needed.

Source reads and lists omit metadata by default. To inspect it, call get_resource with kind: "source" and include_source_metadata: true. source_metadata contains a JSON text chunk, total characters, and next_offset. Pass that offset as metadata_offset and concatenate the chunks before parsing JSON. metadata_limit defaults to 1,000 characters (maximum 4,000). Leave metadata out when polling Source readiness.

Both podcast and episode analytics include analytics_enabled. When false, zero counters mean tracking is disabled, not measured inactivity. complete indicates whether enabled tracking data is complete.

A script request rejected with generation_preflight_rejected has not started generation. Resolve the budget or rate limit, then submit a new idempotency key. Resending the old key repeats that rejection. For uncertain acceptance, retain the original key and inspect the resource; do not submit a new one.

Migration from the previous tools

Replace each previous tool with the call below, and validate its arguments against the newly discovered schema. Existing saved resource IDs remain usable. Old tool names are not aliases on the MCP endpoint.