Small Poke recipes and hosted MCP tools that check the thing, brief context, and draft the next action.
Calendar, incident feed, RSS changes, status endpoint.
Only context that changed since the previous run.
One suggested reply, one next action, clear limits.
Public check returns the current MCP service contract.
Each recipe produces context before it proposes action.
Sensitive integrations start behind explicit scopes.
Poke hosts the recipe. TiredAgents hosts the tool.
Recipe portfolio
The first agents are deliberately compact. They check known sources, summarize change, and hand back a bounded next action.
Reads the sources you already check, detects what changed, and drafts the smallest useful next action.
Collects public status, recent deploy notes, and current symptoms into a short responder note.
A local connector can summarize health and draft a fix list.
Useful automation starts with explicit permissions and boring receipts.
The recipe decides when to check. The hosted MCP tool decides how to gather context, normalize it, and return a draft.
Define the recurring check, schedule, sources, and approval point inside Poke.
Test with public sources first, then add private scopes only when the boundary is clear.
Expose a narrow tool contract, status endpoint, and logs that explain what happened.
TiredAgents is not another place to manage work. It is a thin layer that turns recurring checks into context you can act on.
Check the weekend things again: feeds, logs, status, calendar, media queue, and anything that changed.
Three changes need attention. Subscription alert, deploy note, media index lag. Draft reply is ready.
Trust boundaries
Each concept starts with the smallest safe surface and grows only when the operator can see the credential boundary.
The stack is intentionally boring: recipes, hosted MCP tools, a status endpoint, and local connector paths for private systems.
Public experiments
Follow the first TiredAgents launch, ask about a recipe, or check the current MCP health endpoint.