Braingent

Integrations

Claude Integration

Wire Claude Code (and the broader Claude family) to your Braingent memory repo.

Claude Code reads CLAUDE.md automatically at session start. That single file is the entire integration surface — drop a Braingent-flavored CLAUDE.md in the right place and Claude is on the contract.

Where the file goes

Three valid locations, in order of precedence:

  1. ~/.claude/CLAUDE.md — your global rules, applied to every project.
  2. ~/.claude/projects/<repo-hash>/CLAUDE.md — per-project, project-scoped.
  3. <repo-root>/CLAUDE.md — committed alongside the code, shared with teammates.

Most users put their Braingent block in ~/.claude/CLAUDE.md so every project on their machine inherits the same memory contract.

What goes in it

The starter pack’s CLAUDE.md is ready to use. The key sections:

# Engineering Memory: Braingent

`~/Documents/repos/braingent` is my durable engineering memory across all
repos and AI tools. It holds prior decisions, learnings, tool versions,
repository profiles, ticket history, and code review notes.

**Before planning any non-trivial task:** search Braingent for prior
context. Start at `~/Documents/repos/braingent/CLAUDE.md` for the thin
read order. When MCP tools are available, prefer `braingent_guide()`,
`braingent_find(...)`, and `braingent_get(path, depth='summary')`; use
`depth='full'` only when exact evidence or archived/playbook content
matters.

**After meaningful work:** capture a record. Triggers are PR opened,
ticket closed, code review done, key decision made, reusable learning
surfaced, or any of: "dump this to braingent" / "save to braingent" /
"capture" / "task done thanks" / "ok done".

**Never** store secrets, credentials, tokens, or sensitive personal data
in Braingent.

Replace ~/Documents/repos/braingent with your actual memory repo path.

Claude Code supports MCP. Connect Braingent’s MCP server for clean structured tools.

~/.claude/config.json (or per-project equivalent):

{
  "mcpServers": {
    "braingent": {
      "command": "braingent",
      "args": [
        "mcp", "serve",
        "--path", "/Users/you/Documents/repos/braingent"
      ]
    }
  }
}

Restart Claude Code. The tools braingent_guide, braingent_find, and braingent_get should appear in the tools list.

Smoke test

Open a new Claude Code session in any repo and ask:

What’s the path to my Braingent memory repo, and what’s the first thing you’ll do before planning a non-trivial task?

A correctly wired Claude responds with the absolute path and restates the search-before-plan, capture-after-work contract.

Subagent pattern

When you dispatch Claude subagents (via the Task tool) for parallel work, don’t have each subagent rerun memory lookups. Instead:

  1. The parent agent retrieves a focused context pack once (workflows/retrieve-context.md).
  2. The parent passes that pack into each subagent’s prompt.
  3. Subagents work; parent captures the outcome.

This avoids N round-trips through memory and keeps subagent context focused on the specific subtask.

Common pitfalls

  • CLAUDE.md not loading. Filename is case-sensitive — must be exactly CLAUDE.md. Restart Claude Code after creating or moving the file.
  • Path placeholders not replaced. Smoke-test the absolute path question; if Claude says <MEMORY_PATH>, you skipped step 2.
  • Memory-aware subagents not aware. Make sure your subagent dispatch prompt includes the relevant context pack.

Where to go next