Search

raph search looks up code, docs, and knowledge in the graph from the CLI when an agent does not have the raph MCP server connected. It uses familiar search switches, so agents can get better graph-backed results without learning a new query language. It emits JSON when called by an agent or through a pipe, and compact text in a terminal.

Modes

raph search "database connection"        # ranked keyword (bm25), default
raph search "ResponseWriter" --literal   # exact substring
raph search "Open[A-Z]\w+" --regex        # Go regexp
raph search "config loader" --vector      # semantic graph search (needs provider)

Filters

raph search "auth" --type func --type type   # only functions and types
raph search "TODO" --global                    # across all workspaces
raph search "handler" --path ../other-repo     # scope to a specific workspace
raph search "router" --limit 20

Node types include func, type, file, markdown_chunk, file_chunk, doc, and doc_chunk.

Output

Each match returns the node id, type, name, url (path#symbol), and an excerpt. Default mode is ranked keyword search; --literal performs exact substring search; --regex uses Go regexp; --vector searches indexed graph nodes with embeddings. In JSON mode results come back as a structured matches array ready for an agent to consume.

The same engine is exposed to agents as the MCP search tool when MCP is available — see MCP & CLI.