Anything that indexes is restricted to the server's working directory unless `CGC_ALLOWED_ROOTS` names more, so a stray instruction cannot point it at your home folder. The graph is what the tools are for: dead code and cyclomatic complexity are queries over structure, not text searches, and `execute_cypher_query` is there for the questions nobody wrote a tool for.
A code-graph server. It parses repositories with tree-sitter into a graph database, then exposes that graph to an agent: relationships between symbols, complexity metrics, and raw Cypher when the shaped tools are not enough. The embedded database is the default, so there is nothing to run alongside it.
- Local folders and installed packages indexed into the graph, with repositories listed, removed and counted — `add_code_to_graph`, `add_package_to_graph`, `list_indexed_repositories`, `delete_repository`, `get_repository_stats`
- Keyword search over indexed symbols and content, with fuzzy matching and an edit distance — `find_code`
- Relationship analysis across files, dead code detection, cyclomatic complexity, and the most complex functions ranked — `analyze_code_relationships`, `find_dead_code`, `calculate_cyclomatic_complexity`, `find_most_complex_functions`
- Raw Cypher against the graph, with query visualisation and report generation — `execute_cypher_query`, `visualize_graph_query`, `generate_report`
- Directories watched for changes, so the graph stays current — `watch_directory`, `list_watched_paths`, `unwatch_directory`
- Job control for long indexing runs — `list_jobs`, `check_job_status`
- Java Spring specifics: endpoints, beans and datasource nodes — `find_java_spring_endpoints`, `find_java_spring_beans`, `find_datasource_nodes`
- Multi-project handling: discover indexed contexts in sub-directories and switch the session between them — `discover_codegraph_contexts`, `switch_context`
- Package indexing across python, javascript, typescript, java, c, go, ruby, php and cpp
`pip install codegraphcontext`. The database is embedded by default — FalkorDB Lite, with KuzuDB as a cross-platform alternative and Neo4j available through `codegraphcontext neo4j setup` if you want an external server. Indexing, bundle load, watch and context-switch tools only accept paths under the server's working directory plus whatever you list in `CGC_ALLOWED_ROOTS`, colon-separated on Linux and macOS, semicolon-separated on Windows. A Docker image is published. Version 0.6.5.
One command — pip install codegraphcontext