Model Context Protocol (MCP) API

The DAMS MCP server (dams mcp, stdio) exposes code search, the SQLite cache, and Graphify path retrieval directly to LLM agents — so they use structured tools instead of raw filesystem scans.

Starting the server

dams mcp
# registers in client IDE configs (Cline, Claude Code, Cursor, Continue) via:
dams configure-agents .

Tools

Tool	Description
`search`	Hybrid semantic + regex code search (Chunkhound cAST chunks)
`search_structural`	GritQL structural pattern search / match
`ripple`	Reverse call graph: what depends on a symbol
`graph_query`	Graphify BFS subgraph query (token-efficient)
`read_cache`	Read entries from the SQLite active-code cache
`query_graph`	Graph-first lookup (paired with `graphify-out/graph.json`)

Example: query_graph

{
  "tool": "graph_query",
  "arguments": { "query": "how does auth flow work", "depth": 2 }
}
// returns a ~2k-token subgraph (vs ~123k naive dump)

MCP Interception Proxy

Beyond its own tools, DAMS ships a protocol-level MCP Interception Proxy that wraps any underlying MCP server. When shims are active, native tool calls (grep_search, view_file, list_dir) are intercepted and routed to the DAMS context dispatcher, returning rich Graphify/Chunkhound context instead of bare tool output.

Agent steering invariant

For every configured agent, DAMS injects these rules:

Query-First — if graphify-out/graph.json exists, run graphify query (or query_graph) before grepping.
Context-Aware Scoping — use graphify path / explain for relationships and focused symbols.
AST Update Routine — after editing files, run graphify update . (AST-only, zero token cost).