Code intelligence with Guaardvark (MCP tools)
| tool | use |
|---|---|
list_code_repositories |
which folders are marked as repositories (gives the folder id) |
search_codebase |
semantic search over the indexed project: "where is thinking enabled per model" |
search_code |
case-insensitive regex across files |
list_code_files |
directory listing to orient |
read_code |
a whole file with line numbers |
read_ast_node |
one class or function by name from a Python file |
get_repository_map |
PageRank-ranked architectural map for a folder id |
get_dependency_graph |
file-level import graph for a folder id |
analyze_code |
structure, patterns, improvement notes for a file |
codegen |
a complete modified version of a file from instructions (writes a new file) |
verify_change |
confirm text now exists in a file after an edit |
map_codebase |
run the System Mapper: stats plus ranked findings |
self_improvement_status |
can the self-improvement engine run (lock, flag, already running) |
swarm_status |
the coding swarm (see the swarm skill) |
Pattern
list_code_repositories→ folder id.get_repository_mapfor the shape,search_codebasefor the question.- Read the exact code (
read_code/read_ast_node) before answering; quote file and line. - Edits: make them with your own editor tools in the user's checkout;
verify_changeafterwards.
Rules
- These tools read Guaardvark's index, which can lag the working tree; when it matters, read the file from disk too.
map_codebaseandself_improvement_statusare read-only here. Dispatching a fix to the self-improvement engine is done in the Studio, on purpose.