Skip to content
plainsight.pro

Connect Plainsight Brain to Claude Code

Purpose

Plainsight Brain is our team's shared memory: an MCP server that gives Claude Code your personal memory, the team's lessons, project context, the playbook, our engineering standards, the brand kit and company knowledge. This guide connects Claude Code to it so every session starts with context and ends with a lesson for the next person.

New to the Brain?

Start with 🧠 AI-Native, In Plain Sight: why we built the Brain, what it already does, how the relationship map works, My Brain, and setup for other tools (our AI Assistant, Codex, Cursor, Copilot). This page covers the Claude Code setup in detail.

How It Works

%%{init: { "flowchart": { "useMaxWidth": true } } }%%
flowchart LR
    A[Claude Code] -->|MCP over HTTP| B[Plainsight Brain]
    B --> C[Your memory]
    B --> D[Team lessons]
    B --> E[Playbook and standards]
    B --> F[Company knowledge]
    B --> G[Brand kit]
    style B fill:#2196F3,color:#fff

Claude Code connects to the Brain over MCP. Authentication uses your @plainsight.pro Entra ID: the server advertises its sign-in endpoints, so Claude Code handles the login for you.

Prerequisites

Requirement Details
Claude Code CLI installed (npm install -g @anthropic-ai/claude-code)
Plainsight Entra ID Your @plainsight.pro account
Brain access Your account must be allowed to use the Brain app registration. Ask an admin if you get a login loop.

Step 1: Add the Brain

For every project on your machine (recommended), run this once:

claude mcp add --transport http --scope user plainsight-brain https://barabas.plainsight.pro/mcp

That's it: no tokens, no secrets, no local server to run.

Only for one repository, shared with your team

Use --scope project instead. Claude Code then writes a .mcp.json in the repo root, which you can commit so everyone on the project gets the Brain:

{
  "mcpServers": {
    "plainsight-brain": {
      "type": "http",
      "url": "https://barabas.plainsight.pro/mcp"
    }
  }
}

Already connected at user scope? Then you don't need this as well.

Step 2: Authenticate

  1. Start Claude Code and type /mcp
  2. Select plainsight-brain and choose to authenticate; your browser opens
  3. Sign in with your @plainsight.pro account
  4. Approve the consent prompt (first time only)
  5. Tokens are stored for you, so later sessions reconnect without asking

Run /mcp again to check: plainsight-brain should show as connected.

Step 3: Follow the Agent Lifecycle

Every meaningful session has three moments. They make sure Claude gets context at the start and the team gets a lesson at the end.

%%{init: { "flowchart": { "useMaxWidth": true } } }%%
flowchart TD
    A["brain_start<br/>Project context, your notes,<br/>team warnings, standards"] --> B["Work<br/>brain_expand / brain_search /<br/>recall / playbook_search / ..."]
    B --> C["brain_finish<br/>Session summary plus one lesson,<br/>shared with the team if you say yes"]
    style A fill:#4CAF50,color:#fff
    style C fill:#FF9800,color:#fff

Start every task by telling Claude what you're working on. It calls brain_start, which returns project context, past findings and the prevention cards the team already paid for.

End every task by letting Claude call brain_finish. It stores a short summary and the lesson worth keeping. If a warning from the briefing actually changed what you did, let Claude pass its id in applied_finding_ids: that's the signal that keeps a useful lesson alive.

Without brain_finish, your session is invisible to the team

The work still happens if you skip the finish. The team just doesn't learn anything from it.

Make It Stick in Your Repo

Agents follow what's written down. Put the habit in the repo's CLAUDE.md, and the same lines in an AGENTS.md for colleagues who use Codex, Copilot or Cursor:

## Plainsight Brain
- Start every task with brain_start(project="<repo-name>", task="<what you are doing>").
- End every task with brain_finish, including one durable lesson.
- If the plainsight-brain tools are missing, stop and ask. Don't guess.

Keep secrets and infrastructure IDs out of these files: they get committed.

What You Get

The exact set of tools depends on your role and on which features are switched on. These are the ones you'll use most:

Category Examples What it does
Session lifecycle brain_start, brain_expand, brain_finish Start with context, pull more detail on demand, end with a lesson
Search everything brain_search One question across lessons, your memory, projects, plans, files and the company knowledge graph
Memory remember, recall, correct_memory, archive_memory Your private memory. Corrections keep the old version in the history
Team lessons team_search_findings, team_share_finding What colleagues learned and chose to share
Company knowledge brain_search, search_knowledge_base, my_knowledge_access The relationship map plus our documents, filtered to what you may see
Projects and plans sync_project, get_project_brief, save_plan Project context and plans that survive a context reset
Shared projects prepare_project_share, commit_project_share Hand a project to a colleague, previewed before it's shared
Playbook and standards playbook_search, sdlc_search_guidance, sdlc_check_conformance How we do things here, delivered while you work
Brand brand_get_kit, brand_check The complete brand kit, then a line-by-line check of what you made
Audits store_audit, start_reaudit, compare_audits Record review findings and compare each re-audit with the last one

Admin-only for now

Automated repository review and the secret vault are visible to Brain admins only. For everyone else these tools don't appear at all, which is on purpose.

To see what the Brain keeps for you, open My Brain.

Combining with Other MCP Servers

The Brain works alongside other MCP servers. In a project .mcp.json, add them as siblings under mcpServers:

{
  "mcpServers": {
    "plainsight-brain": {
      "type": "http",
      "url": "https://barabas.plainsight.pro/mcp"
    },
    "databricks": {
      "command": "uv",
      "args": ["run", "--directory", "/path/to/ai-dev-kit", "python", "databricks-mcp-server/run_server.py"],
      "env": { "DATABRICKS_CONFIG_PROFILE": "my-workspace" },
      "defer_loading": true
    }
  }
}

Each server is an independent connection. The Brain uses Entra sign-in in your browser, while Databricks uses a local CLI profile, so they don't interfere with each other. See Connect Databricks MCP to Claude Code.

Who You Are to the Brain

The Brain reads who you are from your sign-in. Almost everyone is in the first row.

Tier Who Access
Personal (default) Any Plainsight sign-in Your memory, projects, plans and history, plus all shared team knowledge. Secrets only if you're an admin.
Team A shared automation token Shared knowledge only. No personal memory. No secrets.
Workspace A workspace token One collaboration space. Nothing else.

Troubleshooting

Symptom Fix
Sign-in loop or it never completes Your Entra account may not have access to the Brain app registration. Ask an admin to add you.
plainsight-brain shows as disconnected in /mcp Run /mcp, select the server and reconnect. The token may have expired.
401 Unauthorized on Brain tools Same: reconnect through /mcp.
Tools appear but calls fail Check /mcp for details. A common cause is Entra consent not yet granted for your account.
Not sure if it's you or the service Open barabas.plainsight.pro/readyz. It needs no login and reports whether the Brain and its stores are reachable.
Server returns 503 The server is starting up. Wait about 30 seconds and retry.
Still connected to an old address If your config points at a *.azurecontainerapps.io URL, replace it with https://barabas.plainsight.pro/mcp and sign in again.