ACR — Agent Composition Records
Register agent interactions and query behavioral lenses through ACR so your copilot remembers how it should act across sessions.
Overview
Tethral ACR MCP is a Build-phase MCP server that logs AI agent interactions and queries behavioral lenses from the Agent Composition Records registry.
What is this MCP server?
- ACR — Agent Composition Records registry via MCP (title in server manifest)
- Log agent interactions to a shared interaction profile store
- Query behavioral lenses for consistent agent tone and constraints
- Optional ACR_API_URL (defaults to https://acr.nfkey.ai)
- npm @tethral/acr-mcp v2.4.1 with stdio transport
- MCP server version 2.4.1
- npm package @tethral/acr-mcp
- Default API https://acr.nfkey.ai
Community signal: 1 GitHub stars.
What problem does it solve?
Solo builders lose consistent agent behavior across threads because interaction history and composition rules live only in ephemeral chat context.
Who is it for?
Builders shipping multi-step agents who want a hosted composition registry without building their own interaction database first.
Skip if: Projects that need only local .md skills with no external API, or teams with strict offline-only agent requirements.
What do I get? / Deliverables
After registration, your agent can write interactions to ACR and pull behavioral lenses so replies stay aligned with how you composed the agent.
- Logged interaction records visible through ACR
- Lens-guided agent behavior queries from MCP tools
- Reusable composition profile across dev and operate phases
Recommended MCP Servers
Journey fit
Agent composition records are created while you build agent stacks and reused when you operate and grow productized assistants. Agent-tooling is the canonical shelf for registries that log interactions and expose composition metadata to MCP clients.
How it compares
Hosted interaction registry MCP, not an in-repo SKILL.md or a generic logging SaaS with no agent lenses.
Common Questions / FAQ
Who is ACR MCP for?
Indie developers and small teams building agent products who want centralized Agent Composition Records and queryable behavioral lenses.
When should I use ACR MCP?
Use it while building agent-tooling and when operating assistants that must recall interaction patterns and lens rules across sessions.
How do I add ACR MCP to my agent?
Install @tethral/acr-mcp, add a stdio MCP entry in Claude Code or Cursor, set ACR_API_URL if not using the default https://acr.nfkey.ai, and restart.