Behavioral scaffolding · Early prototypes
A more considered pace for AI companions.
Paperkinn builds experimental Model Context Protocol (MCP) plugins for Claude-based agentic companions, with ADHD minds in view.
The work explores adaptive pacing, clearer next steps and care boundaries: practical support for the executive-function demands of working with an agent.
The approach
Keep the next step clear.
Leave room to pause.
A capable agent can still give you too much to hold at once: a long answer, several decisions, another branch of work.
Paperkinn explores how behavioral heuristics and MCP state indexing can give a companion more structure. The aim is to keep the task visible while allowing the pace of the conversation to change.
Adaptive pacing
Explore rules for shorter replies and pause checkpoints, informed by session context and user-defined intervals.
Less to hold in mind
Break open-ended work into bounded steps. Keep the current decision and one next action in view.
Care boundaries
Give agreed rest intervals a place in the agent’s workflow, with a clear checkpoint before a pause and a place to resume.
A thread to return to
Use MCP state indexing to make task context, decisions and restart points available across agent runs.
The founder
Psychology informs the architecture.
Paperkinn is founded by Z, an architect of psychologically-aware agent architectures with a Master’s in Community Psychology and a Bachelor’s in Psychology. A New Zealand ASEAN Scholar (NZAS) awardee, Z brings more than eight years of experience in human behavioral systems and university instruction to the design of agentic companions.
Let’s talk about how companions should work.
For questions about the prototypes, technical conversations or potential collaboration, write to Z.
hello@paperkinn.com