About miru
miru provides hybrid semantic and keyword code search for coding agents, returning relevant file paths, line numbers, and source excerpts. It combines embeddings, BM25, and code-aware ranking, indexing on first use and watching local files for changes.
Key features
- 01Search code by meaning
- 02Find strings, regex matches, and filenames
- 03Read neighboring code chunks
- 04Find similar code in other files
- 05Syntax-aware chunking for 26 languages
Use cases
- 01Locate authentication configuration in a codebase
- 02Find a constant or filename
- 03Inspect code around a given location
- 04Identify similar implementations across files
MCP support
miru exposes tools over MCP for search, locating matches, expanding code context, and finding related code. The server builds an index on first use and watches local files for changes.
Configuration
Local server config is the same for every client. Copy it into your client’s MCP config file.
npx -y @takara-ai/miru-codeTAKARA_API_KEYSecretTakara API key. Optional: without it, the MCP auth tool signs in with a device code.Local servers need a runtime on your machine (e.g. Node.js 18+). Config file locations differ by client; see your client’s docs.
Tools
Local servers expose tools only once running on your machine, and the maintainer has not provided a tool list.
Details
From the project READMEMiru (見る)
Hybrid code search for AI coding agents.
Ask where something happens in your codebase. Miru returns relevant file paths, line numbers, and source excerpts through MCP.
Install
Requires Bun 1.1 or later, with bunx on your PATH.
Claude Code (recommended)
Run inside Claude Code:
/plugin marketplace add takara-ai/miru-code
/plugin install miru@miru
Reload plugins or restart Claude Code, then ask:
Use Miru to find where authentication is configured in this project.
If credentials are missing, Claude can use Miru's auth tool to start sign-in. Approve the login code, then retry the search. Plugin credentials are separate from CLI credentials.
Other agents
See agent integrations for Codex and Cursor plugins. The Cursor plugin provides skills and rules; use the CLI installer below to connect its MCP tools.
Install (CLI fallback)
For terminal use or agents configured through the CLI:
bun add -g @takara-ai/miru-code
miru setup
miru install
Choose your agents in the installer, then restart them. See supported integrations. Use miru uninstall to remove the configuration.
Try it
miru search "auth middleware" ./src
miru locate REDIS_HOST ./src
miru expand src/auth.ts 42 ./src
miru find-related src/auth.ts 42 ./src
Use --json for scripts. Run miru <command> -h for options.
MCP tools
| Tool | Purpose |
|---|---|
search | Find code by meaning |
locate | Find exact strings, regex matches, or filenames |
expand | Read neighbouring chunks in the same file |
find_related | Find similar code in other files |
auth | Sign in or configure a SageMaker endpoint |
read_benchmark | Read token-savings totals when benchmark mode is enabled |
The optional repo argument defaults to the server's working directory. Miru builds the index on first use and watches local files for changes.
How it works
Miru combines embeddings, BM25, and code-aware ranking. Tree-sitter provides syntax-aware chunks for 26 languages, with structural and line-based fallbacks for other files.
See chunking and caching and the ranking evaluation.
Privacy and API usage
By default, Miru sends file contents as source chunks and search queries over HTTPS to Takara for embeddings. API usage may incur costs under your Takara plan.
For embeddings on your own infrastructure, use AWS SageMaker. Set MIRU_WORKSPACE_ROOT to restrict local MCP repository paths to one directory.
Documentation
- Reference: CLI commands, MCP parameters, configuration, and benchmarks
- Agent integrations: plugin capabilities, configuration paths, and optional skills
- Contributing: development setup, checks, and releases
Credits
Miru builds on work by MinishLab:
- semble: file walking, indexing, BM25, hybrid search, and ranking signals.
- potion-code and Model2Vec: WordPiece tokenizer.
See NOTICE for attribution and licence text.
License
Comments
New to MCP?
From what the protocol is to installing your first server, these answers cover most first questions.
An open protocol that defines how AI applications connect to external tools and data in a uniform way. Clients talk to the model; MCP servers provide tools, resources and prompt templates. They exchange standard messages, so one server can be reused by many clients.