LemonCrow plugin
LemonCrow adds local repository search to Claude Code, Codex and OpenCode and a browser Reader for reviewing agent changes
- 71 ★ on the repository
- Last commit
Broad access
Why this level
- The plugin gets bash and edit tools and works with repository files, and the installer runs as a script from the internet
- The engine run from source sends anonymous telemetry by default, and Hosted Runtime sends data to LemonCrow infrastructure
Install
- Works in
Manual install
curl -fsSL https://lemoncrow.com/install.sh | bashInstalls the local lc client and a Review server on port 7420, no account needed. A script from the internet: read it first.
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- The plugin gets bash and edit tools and works with repository files, and the installer runs as a script from the internet
- The engine run from source sends anonymous telemetry by default, and Hosted Runtime sends data to LemonCrow infrastructure
This is third-party code. Review the repository files before installing.
What it does
LemonCrow has two products. Runtime attaches to a coding agent as a plugin and gives it a local repository index and a set of tools: code_search finds symbols and strings in the index, read reads ranges, edit makes verified changes, bash runs bounded commands, and relations and graph show relationships and the import map. lc with claude starts Claude Code with the plugin for one launch without touching its settings, model or authentication. For a permanent setup there is lc rt enable with the agent name. The index is incremental and lives in .lemoncrow/workspace/, with no background daemon and no open port. Review covers checking what the agent changed. lc review freezes the current change and opens it in the browser: what changed, which definitions affect code outside the diff, which checks actually ran. Reviewed marks attach to content, so after the agent revises the code only the changed places come back. A plain lc review never calls a model, and AI review is switched on with --ai, a provider key or an agent you are already signed in to (--agent claude). There is a pre-push hook and a GitHub Action for pull requests. Local Runtime Lite is deliberately simple: lexical search with no embeddings. Deeper semantic search goes through the paid Hosted Runtime on LemonCrow infrastructure.
Who it is for. Developers who work through Claude Code, Codex, OpenCode or Cursor and want better repository navigation and calmer review of agent changes.
Good fit when
- In a large repository the agent spends a long time looking for the right place and needs a local index of symbols and imports
- An agent rewrote dozens of files and before committing you want to walk through the change in the browser Reader
- The agent edits code in several rounds and after each one you want to reread only the new changes
- A pre-push hook runs AI review through an already signed-in Claude Code and stops only on a real finding
- Pull requests need comments produced with your own model key, and the GitHub Action does that
Not a fit when
- You need deep semantic search across several repositories: local Lite is lexical and covers a single repository
- You need a verdict such as safe to merge: LemonCrow does not issue one, the decision stays with the human
- You work on Windows: the release is qualified on Linux with Python 3.13, with macOS builds published
- Running scripts from the internet is forbidden: the installer is a curl call to a lemoncrow.com script
Example request
Start Claude Code through lc with, find every place where parseConfig is used and show the import graph around itLimitations
- The release is qualified on Linux with Python 3.13, and macOS builds are published.
- The installer does not configure Cursor: it needs a source checkout and the scripts/install_cursor.sh script.
- Hosted Runtime requires an invited account and the paid Pro plan. Managed execution and cross-workspace context are not available yet.
- Figures in the README (on SWE-bench Verified the cost is 29.5% lower) were measured on published task sets and are not guaranteed for your repository.
- GitHub shows the license as NOASSERTION, while the README states Apache-2.0.
- The engine run from source sends anonymous telemetry by default, and the installed lc client does not.
How to disable. A permanent setup is removed with lc rt disable and the agent name, and a launch through lc with leaves nothing in the agent settings. lc uninstall all lists what it will remove and asks separately about your data, git hooks and the .lemoncrow folder.
FAQ
How do I start Claude Code with LemonCrow?
Install lc and run lc with claude in the Git repository root. The plugin attaches for one launch, and Claude Code settings and authentication stay unchanged. lc rt enable claude makes it permanent.
Does lc review call a model?
No, a plain lc review is deterministic and never calls a model, even with a key in the environment. The --ai flag switches on AI review.
What goes to LemonCrow during local use?
Nothing: sources, the index and tool execution stay on the machine, and no account is needed. The agent still talks to its own model provider as before.
Does LemonCrow work with Cursor?
Yes, but the installer does not configure Cursor: the integration is installed from source with scripts/install_cursor.sh.
How does local Runtime Lite differ from Hosted Runtime?
Lite searches lexically over a local index with no embeddings and covers a single repository. Hosted Runtime adds semantic and hybrid retrieval on LemonCrow infrastructure and needs paid access.
Discussion
No questions or reviews yet.
Related
PluginsA code-review CLI and plugin for agents: reads a git diff and writes line-level comments with context
A Claude Code plugin: Claude does the task, then a stop hook starts an independent Codex review and Claude fixes the findings in the same session
A local semantic code index: the agent searches by meaning, then follows hits into definitions, callers and dependency paths
A skill and CLI for code review: a team of reviewer agents inspects changes independently, debates findings and produces one report
Local code search for agents: finds the files and fragments that matter and keeps context within a set token budget
A Claude Code plugin that runs work through a router, 14 agents and 8 workflows, with state on disk to survive context compaction
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Authentication | Repository stars | Main difference |
|---|---|---|---|---|
| LemonCrow (this card) | no | 71 | ||
| code-index-mcp | no | not required | 162 | an MCP server for a code index without review of agent changes |
| code-review-graph | no | not required | 32,733 | a dependency graph for review, with no Review Reader and no local agent tools |
README in short
The README starts with a quick start: install, a key, reviewing your own change with lc review --ai and attaching Runtime to agents. Next come Runtime Lite with a table of modes, a section on reviewing agent changes with command examples, a table of results on SWE-bench and Terminal-Bench and a diagram of how Lite works. The page ends with supported hosts, a section on privacy and telemetry and a list of limitations.