Podda limits

- Inline MCP artifact: 1 MiB decoded; production connectors pass text arguments around 200 KiB reliably.

- Staged file session (stage_app_files): 128 files plus podda.json, 32 MiB decoded total, one-hour session lifetime; keep each call near 200 KiB of text and base64 entries near 12 KB raw. At most 8 content_url entries per call, each up to 25 MiB fetched server-side.

- Staged ZIP: 32 MiB compressed and 32 MiB decoded.

- Source edits (edit_app_source): at most 50 edits per call; old_string/new_string up to 100 KB each. Single-file reads (read_app_source) cap at 300 KB.

- Chunked upload (action=upload_chunk): at most 640 KiB of raw bytes per chunk and 1000 chunks per upload, sent strictly in order. Keep chunks near 12 KB raw when client argument limits are unknown; base64 arguments stall or truncate in some connector transports.

- Static asset: 25 MiB each; at most 128 app files.

- Client HTML entrypoint: 512 KiB.

- Generated server module: 1 MiB uncompressed.

- KV JSON value: 64 KiB.

- Private file: 10 MiB each, 100 objects and 100 MiB per app.

The staged limits are Podda product boundaries. Cloudflare currently permits larger Worker and asset collections, but larger generated deployments increase validation, startup, abuse, and latency risk.

Last updated 22 August 2026 · Documentation version 7