Vera CLI
A local code search in Rust for agents: BM25, vectors and reranking, 65 languages, offline, with a skill and MCP server.
- 147 ★ on the repository
- Last commit
- MIT
Minimal access
Why this level
- The installer downloads a binary and a model and writes the skill into agent directories.
Install
- Works in
Manual install
bunx @vera-ai/cli installThe first quick start option in the README, followed by `vera setup --potion-code --index .`. `npx -y @vera-ai/cli install` and `uvx vera-ai install` work instead of `bunx`.
Access and checks
How we reviewMinimal accessThe tool runs nothing and makes no network calls.
- The installer downloads a binary and a model and writes the skill into agent directories.
This is third-party code. Review the repository files before installing.
What it does
Vera finds code by behavior and by name. A query like "where expired sessions are cleaned up" returns functions and classes with paths and line ranges, while `parse_config` finds the same function by exact name. Each query combines BM25 with embedding search, then applies ranking that puts definitions above mentions and source above tests and docs. It is a single native binary. The index lives in the project, search runs on your machine, and the default model runs on an ordinary CPU. `vera agent install` adds a skill to more than 30 agent clients, and `vera mcp` starts an MCP server with seven tools, among them `search_code`, `find_references` and `structural_search`.
Who it is for. Developers whose agents burn tokens reading extra files in large repositories and must not send code to an external index.
Good fit when
- The agent cannot find the right function in a monorepo from a behavior description.
- Before a refactor you need to see who calls a symbol and which functions have no callers.
- Code must not leave the machine, yet search is required.
- One query searches both the project and the source of dependencies from `~/.cargo/registry`.
Not a fit when
- The repository is tiny and plain `grep` is enough.
- Search over documents, chats or a knowledge base is needed rather than code.
- The environment cannot download the binary and model, so a manual install from the docs is required.
Example request
Use Vera to find where session expiry is handled in this project and show the functions with paths and callers.Limitations
The first run downloads a native binary, an embedding model and ONNX Runtime, so network access or a manual install is needed. 54 of the 65 languages get symbol-level chunks, and the rest are split by section or line windows. A large tree needs a separate indexing pass, and `--full-text-only` skips the model and searches with BM25 alone. Linux on aarch64 needs glibc because no musl build exists for it. Speed and quality figures in the README were measured by the author on their own machine.
How to disable. Remove the skill with the installer commands or by hand from agent directories, delete the `vera` entry from MCP configuration and remove the binary. The index lives in the project and is deleted with its index folder.
FAQ
Should I connect the skill or MCP?
The README recommends the CLI plus the skill. The skill tells the agent when to use `vera search`, `vera references` or plain `grep`. The `vera mcp` server is for clients that prefer it.
Does code go to an external service?
Not by default: indexing, storage and search run on your machine, and the model runs on the CPU with no API key. An external embedding backend is enabled separately.
How do I refresh the index?
`vera update .` re-indexes only changed files, and `vera watch .` does it continuously.
How do I search dependencies?
Run `vera index ~/.cargo/registry/src --full-text-only` once, then a query with `--also ~/.cargo/registry/src` merges the lists.
Which agents get the skill?
The installer supports more than 30 clients, including Claude Code, Codex, Cursor, OpenCode, Gemini CLI and Copilot. For others `npx skills add VeraTools/Vera` works.
Discussion
No questions or reviews yet.
Related
CLI toolsLocal search and MCP server for agents: one Rust binary that indexes code, docs and logs
codesearch searches code across several repositories with hybrid retrieval and runs fully offline, without a GPU or Docker.
Source search ranks matches using code structure and gives agents relevant snippets without a separate search index.
An MCP server that indexes a local repository with tree-sitter and gives the agent code search, file lookup and a structure summary of a file
A structural code search MCP server for agents: finds the right file by query intent and runs locally
Vexor: semantic search over files and code with an MCP server, where an agent finds a file by what it contains.
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Authentication | Repository stars | Main difference |
|---|---|---|---|---|
| Vera (this card) | no | 147 | ||
| semble | yes | not required | 6,404 | also searches code locally with no keys; Vera combines BM25 with embeddings and also answers structural questions: definitions, callers, functions with no callers |
| Serena | no | not required | 31,335 | works through language servers and edits code by symbol; Vera only searches an index and changes no files |
README in short
The README covers the quick start, how search works, setting up the skill and the MCP server, a comparison with other local tools and links to detailed documentation pages.