Overview

Available today

The Claude Code connector is Meetless Agent’s first connector and the product that ships today. It takes the model from How it works and lands it on a local Claude Code setup. If you read one section as a developer, read this one and the pages under it.

What the connector is

The connector is three surfaces working together, driven by the mla CLI:

• Capture hooks. Local hooks observe the session (prompts, decisions, tool calls, produced documents) and feed them into governed memory as evidence. This is Capture in practice.
• The MCP server. mla mcp runs a Meetless MCP server over stdio, authenticated as you and scoped to this workspace. It is how the agent queries governed memory from inside a session, and how Ask works in-session.
• The /mla skill. A skill that drives the higher-level flows, onboarding a repo and reviewing knowledge, from inside the agent.

The model, recapped in one paragraph

Governed memory holds your team’s cited decisions and rules. Confirmed decisions are injected into context on every prompt. You can ask the corpus and get grounded, cited answers. The session’s own output is captured as evidence, born pending, and only a human promotion makes it trusted. Contradictions against the approved corpus are surfaced for a human. One enforcement rule can block on the wire today; broader enforcement is observe-first. Each of those links back to its concept page; this connector does not redefine them, it delivers them.

Note

The concepts are taught once, agent-agnostic, in How it works. These connector pages assume you have skimmed them and focus on what actually happens on Claude Code.

The rest of this section

• Install & activate: get the CLI, sign in, bind a repo, verify the wiring.
• What happens in a session: the lifecycle of a governed session.
• Onboarding a repo: go from an empty workspace to a reviewed starting corpus.
• Reviewing & promoting knowledge: the human gate that turns evidence into trusted knowledge.
• Asking & enforcing: query governed memory and see the enforcement pilot in action.

Connector number one is Claude Code. The model is agent-agnostic, so more connectors can follow; see the connector vision.