Guided first use

Try ops-brain MCP

Self-hosted team bus for AI agents: cross-machine handoffs, check-in, and shared knowledge on your own PostgreSQL. Reads keep Guard's usual Follow the three steps below with your own setup — nothing on this page runs on your machine or changes a policy.

Step 1 · Check fit

What this coverage governs

Self-hosted team bus for AI agents: cross-machine handoffs, check-in, and shared knowledge on your own PostgreSQL. Reads keep Guard's usual handling; writes to the shared bus and both deletes are gated on review. Off until you turn it on.

External — off until you opt in

This coverage stays off until you explicitly enable it from your Guard workspace controls. Reading this page never changes your policy; activation happens in Guard itself.

Harness support

Not documented yet

Contributor-supplied harness support text is not independently verified. Any MCP harness that can launch a local stdio server. The author has run ops-brain with Claude Code, Codex, Grok, Antigravity CLI (Gemini) and Grok Bot.

  • Any MCP harness that can launch a local stdio server. The author has run ops-brain with Claude Code, Codex, Grok, Antigravity CLI (Gemini) and Grok Bot.
  • Linux and macOS, amd64 and arm64 (multi-arch image ghcr.io/thek3nsai/ops-brain, plus a static binary).
  • This Guard entry matches only the stdio command `ops-brain`. Self-hosted HTTP deployments (streamable HTTP with per-agent bearer tokens) aren't matched; set those up on Guard's Protect page.

Prerequisites

Not documented yet

Contributor-supplied prerequisite text is not independently verified. PostgreSQL with the pgvector extension (the project uses pgvector/pgvector:pg18).

  • PostgreSQL with the pgvector extension (the project uses pgvector/pgvector:pg18).
  • DATABASE_URL pointing at that database. ops-brain runs its migrations on start.
  • Optional: an embeddings provider for hybrid search. Without it, search falls back to full-text.

Known limitations

Reviewed as supported

Limitations are copied from the reviewed catalog facts. Contributor-supplied limitation notes are not independently verified. Coverage limits declared in the reviewed catalog appear automatically when this guide is published.

  • Coverage is limited to the reviewed operations and the surrounding Guard policy.
  • A maintainer profile is not a security certification or an official upstream endorsement.
  • Coverage is limited to the reviewed operations and the surrounding Guard policy.
  • A maintainer profile is not a security certification or an official upstream endorsement.

Catalog version 1.0.0 · source snapshot 4023fb6ea9

Step 2 · Set up

Install and activate

Install Guard Open Guard activation

New to Guard? Start with Install Guard. Already installed? Skip install — this coverage stays off until you enable it from your workspace controls.

Activation happens in your Guard workspace. This page links to it; it never enables the coverage for you.

Install

Contributor-supplied install guidance is not independently verified because an exact reviewed argv contract is unavailable. Get ops-brain: the release binary from https://github.com/TheK3nsai/ops-brain/releases or `docker pull ghcr.io/thek3nsai/ops-brain:v6.3.0`.

  1. 1

    Get ops-brain: the release binary from https://github.com/TheK3nsai/ops-brain/releases or `docker pull ghcr.io/thek3nsai/ops-brain:v6.3.0`.

    Self-reported guidance — not verified against a reviewed source.

  2. 2

    Register it in your harness as a stdio MCP server: command `ops-brain`, env DATABASE_URL.

    Self-reported guidance — not verified against a reviewed source.

  3. 3

    In HOL Guard, turn on the mcp.ops-brain entry (opt-in, off by default).

    Self-reported guidance — not verified against a reviewed source.

Activation

No activation procedure was supplied in the reviewed source.

Back to fit

Step 3 · Verify the result

Run a safe test and compare

Safe fixture inputs

Contributor-supplied safe fixture input guidance is not independently verified because an exact reviewed argv contract is unavailable. Use a throwaway database for the first run.

  1. 1

    Use a throwaway database for the first run.

    Self-reported guidance — not verified against a reviewed source.

  2. 2

    Ask the agent to call check_in. It's read-only, so Guard lets it through with normal handling.

    Self-reported guidance — not verified against a reviewed source.

  3. 3

    Ask it to create_handoff addressed to itself. Guard should stop it for review. Approve it, then call list_handoffs to see it.

    Self-reported guidance — not verified against a reviewed source.

  4. 4

    Ask it to delete_handoff on that test handoff. Guard should stop it again. If you approve, the row is deleted for good and an audit_log record is written.

    Self-reported guidance — not verified against a reviewed source.

Expected outcome

Not documented yet

Self-hosted team bus for AI agents: cross-machine handoffs, check-in, and shared knowledge on your own PostgreSQL. Reads keep Guard's usual handling; writes to the shared bus and both deletes are gated on review. Off until you turn it on.

Did it work?

Sign in to record an honest self-report. It is stored as your report — never as a verified evaluation.

Sign in to report

Turn it off or recover

Not documented yet

No disable/remove procedure was supplied in the reviewed source.

Back to setup
Back to coverage details