ARPUseful agents need more than a longer prompt.
Tools, layered context, skill discovery, and an editable system each have a job to do.
Context answers only part of the question.
An agent needs to understand the task, but it also needs a way to act, discover the right procedure, and carry work forward. Adding more information does not by itself supply those capabilities. Morning’s architecture combines purpose-built MCP tools, multiple layers of context handling, skill discovery, and an interface agents can edit and manage.
Tools make actions specific.
A well-defined tool tells an agent what action is available, what inputs it needs, and what result it returns. Permission is a separate question. Connecting a channel should make both capability and authority clearer. In a social workflow, drafting content, reviewing it, publishing it, and verifying the outcome are different steps with different implications.
Layers and indexes help find what matters.
Business records, the current task, a relevant procedure, and provider-specific details should not all compete in one undifferentiated block. Layered handling helps bring the right material into the work at the right point. Skill indexing addresses discovery when a plugin has many capabilities: the system should help the agent find the relevant procedure without treating every skill as equally relevant.
An editable system opens a second kind of work.
Agents can help maintain and adapt the system itself: update a representation, propose a procedure, or connect related records. Changes need the right permissions, checks, and recovery path. This architecture is our direction for more dependable work; it is not a published measurement of how much more consistent every agent will become. That claim needs evaluation in the environment where the system is actually used.
Social MCPs
Give agents specific tools for social workflows, with business context, permissions, and a clear connection to the work.
Meet Social MCPs