1.8 KiB
Codebase Memory MCP
Use codebase-memory-mcp as the primary tool for understanding code structure, symbols, dependencies, call paths, and architecture.
Project selection
-
Indexed project names are stable and equal to repository directory names.
-
Current mappings:
~/code/es/trustyca→trustyca
-
Determine the current repository from its Git root and use its basename as the project name.
-
Call
list_projectsonce per session to verify that the expected project exists. -
Always pass
project="<repo-name>"explicitly to every project-scoped MCP tool. Never rely on automatic project detection, filesystem paths, or a default project. -
Never substitute the workspace path, symlink-resolved path, or parent directory for the indexed project name.
-
If the expected name is absent from
list_projects, do not guess or silently reindex. Report the mismatch and ask whether the repository should be indexed.
Tool usage
- Start architecture exploration with
get_graph_schema(project=...)andget_architecture(project=...). - Use
search_graph(project=...)to locate symbols before callingget_code_snippetortrace_path. - Use qualified names returned by
search_graph; do not construct them manually. - Use
search_code(project=...)for indexed text searches and filesystem search only when graph/index data is insufficient. - Use
detect_changes(project=...)when evaluating the impact of Git changes. - Before trusting an empty result, verify the selected project and indexing status.
- Do not run
index_repository,delete_project, ormanage_adrunless the user explicitly requests an indexing, deletion, or ADR operation.
Example:
list_projects()
search_graph(project="trustyca", name_pattern=".*Certificate.*", label="Function")
get_architecture(project="trustyca", aspects=["all"])