0.1.4

---

One MCP tool that gives your coding agent fast, structure-aware sight into a codebase — and a safe, precise way to change it. Warm, resident, always fresh. Fully local. This package downloads the matching prebuilt vyer binary for your platform on install.

Use it

Vyer is an MCP server — your agent host launches it and talks to it over stdio. You don't run vyer serve by hand; you point your host (Claude Code / Cursor / Windsurf / Cline) at it:

// .mcp.json (or your host's MCP config)
{
  "mcpServers": {
    "vyer": { "command": "npx", "args": ["-y", "@0x1labs/vyer", "serve", "--root", ".", "--watch", "--allow-writes"] }
  }
}

Claude Code one-liner:

claude mcp add vyer -- npx -y @0x1labs/vyer serve --root . --watch --allow-writes

--allow-writes lets Vyer edit (drop it for read-only). Then tell your agent to prefer it:

npx -y @0x1labs/vyer init      # drops a managed note into ./CLAUDE.md (idempotent)

Want a global vyer command? npm install -g @0x1labs/vyer. To check it from a terminal (no agent), use vyer version or vyer query "<search>" — serve only does something when an agent drives it.

What you get

One loop, not just search

// find — describe it or name it
{ "queries": [{ "q": "where do we validate the auth token" }] }

// understand — definition + callers + callees + tests, in one call
{ "queries": [{ "q": "validateToken", "detail": "context" }] }

// change — a surgical, atomic edit (this is code_apply)
{ "edits": [{ "locator": "src/auth/token.rs#validate_token", "rename": "verify_token" }] }

Full documentation

Quickstart, the agent playbook, security model, and benchmarks all live in the repo:

→ https://github.com/chirayuoli/vyer

No prebuilt binary for your platform? Build from source (needs Rust 1.85+): cargo install --git https://github.com/chirayuoli/vyer vyer-server