Persis
Add biologically inspired agent memory—record, recall, reinforcement, decay, clustering, and eight-drive behavior—to long-running Claude or Cursor workflows via a hosted MCP API.
Overview
Persis is a MCP server for the Build phase that provides biological agent memory through fifteen lucian_* tools including recall, reinforcement, eight-drive behavior, and clustering over streamable HTTP.
What is this MCP server?
- Fifteen MCP tools including lucian_record, lucian_recall, lucian_find_similar, lucian_reinforce, lucian_invalidate, and
- Eight-drive behavior model with lucian_drives, lucian_apply_drives, and lucian_tick for timed cognitive updates
- Clustering and consolidation via lucian_clusters, lucian_consolidate, and lucian_maintenance
- Remote streamable-http Persis API on Cloudflare Workers; beta free tier at 60 requests per minute per IP
- Biological-memory metaphors: pattern learning, decay, and reinforcement—not a flat key-value scratchpad
- 15 MCP tools listed (lucian_record through lucian_export)
- 8-drive behavior model documented in server description and tool names
- Rate limit 60 requests per minute per IP during beta; server version on registry stream
What problem does it solve?
Long-running agents lose nuanced patterns and cannot prioritize what to remember, forget, or consolidate without a dedicated cognitive memory layer.
Who is it for?
Advanced solo builders prototyping persistent agents who want decay, clustering, and drive modeling without building cognitive memory from scratch.
Skip if: Beginners seeking a simple local key-value memory, teams needing guaranteed offline-only storage, or products that only need static documentation retrieval.
What do I get? / Deliverables
After connecting the Persis remote MCP, your agent can record, recall, reinforce, cluster, and maintain memories with drive-aware behavior across sessions.
- Record/recall/reinforce memory loop exposed as fifteen MCP tools
- Drive-aware agent behavior and exportable memory stats
- Clustering and consolidation operations without custom cognitive code
Recommended MCP Servers
Journey fit
Persis is adopted when you are composing agent stacks and memory policies during product build; it keeps session learning available across later operate and iterate work but first lands as agent infrastructure. Agent-tooling is correct because fifteen lucian_* MCP tools manage cognitive memory mechanics (drives, consolidate, maintenance) rather than shipping user-facing frontend code or launch distribution.
How it compares
Hosted cognitive-memory MCP with 15 tools and 8-drive behavior, not Konstruct’s concept graph or EvolutionDB’s self-hosted SQL memory.
Common Questions / FAQ
Who is Persis for?
Persis is for developers shipping agent-native products who need pattern learning, memory decay, reinforcement, and drive-aware behavior exposed through MCP.
When should I use Persis?
Use it during Build agent-tooling when you are designing persistent agents that must recall, consolidate, and maintain memories over many sessions.
How do I add Persis to my agent?
Configure the streamable-http remote MCP URL https://persis-api.aerynw.workers.dev/mcp in your client (including any documented headers), then invoke tools such as lucian_record and lucian_recall from the agent.