> For the complete documentation index, see [llms.txt](https://docs.loci-dev.net/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.loci-dev.net/loci-plugin-for-ai-coding-assistants.md). # LOCI Plugin for AI Coding Assistants ## LOCI plugin for Claude Code > One plugin bundles every LOCI skill. Claude Code calls them when it judges useful — at `/plan`, write, review or merge. **We don't enforce; Claude co-reasons.** LOCI ships as a single Claude Code plugin. It registers all of LOCI's skills (plan review, write-time evidence, PR review, merge guard) plus a local MCP server that Claude Code uses to fetch execution signals at runtime. The user installs the plugin once. From that point on, Claude Code decides when to consult LOCI — Aurora Labs does not force the agent to call it. The plugin is the pipe; co-reasoning is Claude's call. *** ### What gets installed A single bundle that contains: * **All LOCI skills** — `/plan` review, write-time signals, PR review, merge guard. Claude Code can invoke any of them. * **LOCI MCP server** — local process Claude Code talks to over MCP to fetch the five execution signals (timing, stack, power, throughput, risk). * **CLI** — `loci auth`, `loci status`, `loci link` for setup and inspection. No source-code instrumentation. No runtime. The MCP server reads compiled binaries (ELF / Mach-O / PTX / Wasm / bytecode) and returns predictions from LCLM, a small specialized model trained on five years of real CPU + GPU execution traces. *** ### Install flow Five steps. Each step has one or two commands; copy and run. #### 1. Download the LOCI plugin One plugin, all skills. Two commands and you're done — Claude Code decides when to call LOCI at `/plan`, write, review, or merge. ```bash # Download the LOCI plugin bundle (all skills + MCP server) $ curl -sSL https://loci.auroralabs.com/install.sh | sh # Register the plugin with Claude Code $ loci link --claude-code ``` #### 2. Add LOCI in Claude Code → `/plugin` Open Claude Code, run `/plugin` and add LOCI from the Marketplace tab. ```bash # Inside Claude Code, open the plugin marketplace $ /plugin > marketplace · search "LOCI" · click Add to Claude Code ``` #### 3. Authenticate to LOCI Browser opens, you sign in, the CLI receives the token. Free tier works on one repo, no credit card. ```bash $ loci auth login > opens browser → sign in → token stored at ~/.loci/credentials ``` #### 4. Enable the LOCI MCP server MCP is how Claude Code talks to LOCI at runtime. Enable it once — Claude Code will co-reason with LOCI from the next session. ```bash # Enable the LOCI MCP server inside Claude Code $ /mcp enable loci > mcp · loci · enabled · 5 signals available (timing · stack · power · throughput · risk) ``` #### 5. You're done — Claude Code co-reasons on its own From here on, Claude Code calls LOCI when it judges it useful: planning a refactor, writing a hot path, reviewing a diff. We don't enforce — **Claude decides.** ```bash # Verify Claude Code is talking to LOCI $ loci status > claude-code · connected · mcp · loci ready · co-reasoning on ``` *** ### How co-reasoning works after install Once the plugin is registered, authenticated, and the MCP server is enabled, Claude Code has LOCI's five signals available the same way it has any other MCP tool. Aurora Labs does **not** add gates, hooks, or hard policy. The agent is in charge: * If Claude Code is planning a refactor that smells performance-critical, it may call `loci.plan_review` against the binary before generating code. * If Claude Code is writing a hot path, it may call `loci.execution_data` after each iteration to see timing / stack / power. * If Claude Code is reviewing a diff, it may call `loci.pr_review` to fetch a verdict + evidence. * If Claude Code judges that the task doesn't benefit from execution signals, it won't call LOCI. That's fine. Every call returns numeric, physically meaningful values (ms, watts, instructions/cycle) bounded by what LCLM has measured on real hardware — no hallucination, by design. *** ### Why a plugin, not a wrapper LOCI doesn't wrap Claude Code or sit between the user and the agent. It exposes itself as **one MCP server + a set of skills**, both registered through Claude Code's native plugin system. Consequences: * **No vendor lock-in on the agent.** Same plugin pattern works for Cursor, Copilot, and 15+ other agents via `skills.sh`. Listed on AWS Marketplace. * **No enforcement layer.** Claude Code is the decision-maker; LOCI is a tool it can pick up. * **No instrumentation.** LOCI reads compiled binaries. Source code never leaves the developer's machine. * **Predictable cost.** Verdicts are capped per plan, never billed per call. *** ### CI path (for completeness) The plugin flow above is for the developer's loop. For PR gating, install the **LOCI GitHub App + Action** — separate path, same engine. See `Quick Start → GitHub · CI + App` on `/product#quick-start`. *** ### Free tier and pricing * **Free** — one repository, no credit card. Works with Claude Code \[following with Cursor, Copilot]. * **Team — $199/mo** — unlimited repos, verdict + evidence on Plan / Write / PR / Merge, 1-business-day support. * **Enterprise — Custom** — self-hosted / air-gapped, SSO/SAML/RBAC, ISO 27001 + ASPICE / ISO 26262 alignment, 120+ patent indemnification, dedicated SE. Pricing details: `auroralabs.com/pricing`. *** ### Reference * Quick Start (web): `auroralabs.com/product#quick-start` * Talk to us: `auroralabs.com/contact` * LOCI install endpoint: `https://loci.auroralabs.com/install.sh` --- # Agent Instructions This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com. ## Querying This Documentation If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question. Perform an HTTP GET request on the current page URL with the `ask` query parameter: ``` GET https://docs.loci-dev.net/loci-plugin-for-ai-coding-assistants.md?ask= ``` The question should be specific, self-contained, and written in natural language. The response will contain a direct answer to the question and relevant excerpts and sources from the documentation. Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.