Agence
Wire policy enforcement, governed skills, shared memory, and multi-LLM consensus into Claude Code or Cursor without bolting on ad hoc guardrails.
Overview
io.github.l-agence/agence is a MCP server for the Ship phase that enforces AI governance policies, skills, memory, and multi-LLM consensus inside your agent toolchain.
What is this MCP server?
- Policy enforcement MCP tools so agents stay inside your allowed actions and data boundaries
- Governed skills and memory surfaced through stdio npm package @l-agence/mcp-server-agence
- Multi-LLM consensus flows for decisions that should not rely on a single model answer
- AGENCE_ROOT environment variable points at your agence install root when not auto-detected from npm
- stdio transport fits local Claude Desktop, Cursor, and Claude Code MCP configs
- server.json version 1.0.0
- npm identifier @l-agence/mcp-server-agence
Community signal: 3 GitHub stars.
What problem does it solve?
Coding agents can drift, contradict each other, and ignore informal rules unless policy and memory sit in a protocol the agent can actually call.
Who is it for?
Solo builders shipping agent-heavy products who want npm-based stdio governance without writing a custom policy microservice.
Skip if: Teams that only need a single-model chat with no enforcement, memory, or multi-model agreement requirements.
What do I get? / Deliverables
After registration, your agent can enforce policies, use governed skills and memory, and run consensus across LLMs through one stdio MCP connection.
- stdio MCP connection to policy, skills, memory, and consensus tools
- Central governance layer callable from agent sessions
- Configurable agence root for local policy and skill assets
Recommended MCP Servers
Journey fit
Governance and policy enforcement belong on the Ship phase shelf because they harden how agents behave before and during production use. Security is the canonical subphase for MCP servers that enforce rules, audit agent actions, and align multiple models under one policy layer.
How it compares
MCP governance integration, not a standalone agent skill or a generic skills.sh catalog entry.
Common Questions / FAQ
Who is io.github.l-agence/agence for?
Indie and solo developers who use Claude Code, Cursor, or similar hosts and want policy enforcement, memory, and multi-LLM consensus exposed as MCP tools.
When should I use io.github.l-agence/agence?
Use it during Ship and Operate when agent behavior must follow explicit policies and when important decisions benefit from consensus across more than one model.
How do I add io.github.l-agence/agence to my agent?
Add the @l-agence/mcp-server-agence stdio server to your MCP client config, set AGENCE_ROOT if needed, and restart the host so governance tools load.