Skip to main content

Using MCP

The assistant-first door: use Vyasa through your coding assistant instead of typing commands — governance as tools your agent calls. (Tool-level detail: MCP Tools.)

The idea in one paragraph​

Your assistant already writes and edits. With Vyasa's MCP server connected, it can also open governed sittings, check gates, record notes, and file work for approval — the same loop as the terminal, invoked as tool calls inside its normal flow. You keep approving at gates; the assistant handles the mechanics between them.

Start​

vyasa mcp

Point your assistant's MCP settings at that server (stdio). It discovers ~90 tools: sessions, nodes, gates, reviews, scores, violations, governance reads, and more — each mirroring a CLI capability, none duplicating behavior.

How a governed task flows via MCP​

  1. You ask your assistant for a change on a known work item.
  2. It opens a session (skill, mode, stage, target — same eight details, as tool arguments).
  3. It produces output into staging and files it for review.
  4. You review the diff and approve the gate — in terminal or console, your choice.
  5. It sees approval, continues to the next item.

Human decisions stay human; machine mechanics fade into tool calls. If a gate rejects, the assistant gets the scorecard back and revises — same loop, no new concepts.

Know the boundary​

MCP tools operate the platform (sessions, gates, nodes). The workflow-defined MCPs are something else: outside services your workflow allows workers to call mid-task. Same letters, opposite sides of the counter — your workflow's list scopes what the assistant may reach.