Measure a server's MCP Tasks durability instead of trusting its declarations.
MCP Tasks promise a durable handle for long-running work, but compliance is not durability: a server on an in-memory backend passes conformance and loses every task on a routine restart, a crashed client leaves expensive work running unseen, and the protocol does not even require idempotent creation. This product measures a server's durability instead of trusting its declarations - create, TTL declaration, TTL honesty re-read inside the window, reconnect, retry-dedup, era-tagged rediscovery, cancel support verified by observation - and publishes DURABLE / FRAGILE / ORPHAN-RISK per server, with a fleet task ledger that inventories every taskId across your servers and a reaper that cancels confirmed orphans with a receipt and reports the un-cancellable ones with their burn instead of dropping them silently.The API verdict is reconnect-only and UNMEASURED-RESTART is provisional; restart legs run only through an operator-supplied restart callback against the operator's own staging/shadow or scheduled instances.
Prices: Per-server durability verdict $0.25/server/mo (x402) · Fleet task ledger + orphan reaper $19/mo (up to 50 servers) · Registry / platform verdict feed $99/mo. First server's first verdict free, one time per operator.
API: https://textforagents.com/products/mcp-task-durability-orphan-reaper/ · Support: support@textforagents.com