# LBrain coding-agents Install LBrain where the agent already lives. The free name is generic. Make it permanent: https://lbrain.ai/claim.html Native MCP + companion skill + fail-closed session-start for Claude Code, Codex, Cursor, Copilot, Grok Build, Gemini CLI, Antigravity, and OpenClaw. This package does **not** modify the LBrain engine. It wires hosts to `lbrain mcp`. The Official MCP Registry / VS Code `@mcp` gallery entry also execs that same stdio server. VS Code: search **LBrain** in Extensions, or `@mcp lbrain`. Marketplace: https://marketplace.visualstudio.com/items?itemName=metavolve-labs.lbrain ## Install the engine first ```bash pip install "lbrain[local]" lbrain init --source ~/notes lbrain import && lbrain embed --stale ``` ## Wire your coding agents ```bash pip install lbrain-coding-agents python3 -m lbrain_agents status python3 -m lbrain_agents install # lists detections, changes nothing python3 -m lbrain_agents install all # every detected harness python3 -m lbrain_agents install claude-code grok-build python3 -m lbrain_agents install all --dry-run ``` A bare `install` is a no-op. That is deliberate: wiring every agent on the machine never happens by accident. ## Paste to a desktop AI (no terminal) Some people never open a shell. They paste this into Claude, Cursor, Codex, Copilot, or Grok and the agent does the work: ``` Install LBrain as my local memory. pip install "lbrain[local]" lbrain-coding-agents If python3 -m lbrain_agents works here, run: python3 -m lbrain_agents install all Otherwise add an MCP server named lbrain: command lbrain, args ["mcp"], stdio. Then call whoami. Prefer binds. Near-miss is not an answer. Abstain if nothing binds. SUPERSEDED must not govern. Leave LBRAIN_HOME alone if it is already set. https://lbrain.ai/integrations.html ``` Existing `lbrain` MCP entries are left untouched unless you pass `--force`. Skills are copied to each host's skills directory. Session-start injects `lbrain whoami` (Claude Code). ## Two-line wrap ```python from openai import OpenAI from lbrain_agents.wrap import wrap_openai client = wrap_openai(OpenAI(), home="~/.lbrain") ``` Recall first. If nothing binds, the model is told to abstain. Writes are opt-in (`remember=True`). ## Agent Plugin `lbrain_agents/share/agent-plugin/` is a portable bundle (`plugin.json` + `mcp.json` + skill) for hosts that speak the Agent Plugins spec. ## Cookbooks 1. Abstain when the record does not bind 2. SUPERSEDED does not govern 3. Model swap, same mind See `lbrain_agents/share/cookbooks/`. ## Safety - Merges only the `lbrain` key. Other MCP servers stay. - Backs up a config once as `*.lbrain-backup` before the first write. - Does not set `LBRAIN_HOME` unless you pass `--home`. A working seat mount is not overwritten. - HTTP MCP is **not** enabled by this installer. stdio only. The engine's HTTP transport has no auth; do not bind it to the world. ## What this package is not This is the free installer and the serve-boundary skill. It is not LBrain Connect, Managed LBrain, or LBrain Govern. It does not issue `gcx://` names, inscribe permanence, or ship study gold sets. The engine's resolver client is open; registrar issuance is not in this tree. ## License BSD-3-Clause. Metavolve Labs, Inc. Patents pending. The licence covers the code; it does not grant patent rights.