codesift

IntelliJ-grade code intelligence for agents — persistent structural + semantic indexing, exposed via CLI and MCP.

Rust-based indexing engine for AI agents and CI. Structural index, MCP agent frontend, and CLI ops/debug implemented; semantic search next.

Development

Uses mise for the Rust toolchain (see mise.toml):

mise trust
mise run fix    # cargo fmt + clippy --fix
mise run check  # CI-equivalent
mise run cli index .
mise run mcp          # MCP stdio server (Cursor / OpenCode)

Agents → MCP (find_references, get_callers, …). Config: .cursor/mcp.json, opencode.json — both use mise run mcp. Humans/CI → CLI (index, status, export).

Dogfooding

Agent exercises with ground truth and baselines: docs/dogfooding/.

Quick links

• Vision — problem, solution, principles
• Documentation hub — reading paths by role
• Capabilities — feature coverage matrix
• Specs — implementation contracts
• Contributing

Interfaces

# Agents — MCP (primary)
codesift --workspace . mcp

# Humans / CI — CLI ops and debug
codesift index .
codesift status
codesift refs --name parse_query --format ascii
codesift export --type ref

License

Apache-2.0. See LICENSE.