Cheat sheet

One page, the daily moves: build one analysis object, then query it: the same interface for Java and Python. For the why and the walkthroughs, see What is CLDK?, Core concepts, and the Quickstart.

Install

pip install cldk

Reachability uses networkx (a CLDK dependency, but install it explicitly if you query graphs directly):

pip install cldk networkx

Construct

One pattern for every language: CLDK(language=...).analysis(...). The analysis level decides how much gets computed.

Java

from cldk import CLDK
from cldk.analysis import AnalysisLevel

# Default level is symbol_table. Use call_graph for callers/callees/reachability.
analysis = CLDK(language="java").analysis(
    project_path="commons-cli",
    analysis_level=AnalysisLevel.call_graph,
)
print(len(analysis.get_classes()), "classes")  # -> 23 classes

Python

from cldk import CLDK
from cldk.analysis import AnalysisLevel

analysis = CLDK(language="python").analysis(
    project_path="my_pkg",
    analysis_level=AnalysisLevel.call_graph,
)
print(len(analysis.get_classes()), "classes")

Argument	Value	Notes
language	"java" / "python"	Picks the analysis facade.
project_path	path to checkout	The project to analyze.
analysis_level	AnalysisLevel.symbol_table (default) / AnalysisLevel.call_graph	Call graph, callers, callees, and reachability need call_graph.

Symbol table

The always-available layer: every class, method, and field, typed and queryable.

Method	Returns	What it gives you
get_symbol_table()	Dict[str, JCompilationUnit | PyModule]	The whole project, file by file.
get_classes()	Dict[str, JType | PyClass]	Every class/type, keyed by qualified name.
get_methods()	Dict	Every method (Java).
get_method(qualified_class_name, qualified_method_name)	JCallable | PyCallable | None	One method; .code is its source body.

m = analysis.get_method(
    "org.apache.commons.cli.OptionBuilder", "create(String)"
)
print(m.code)  # -> source body of OptionBuilder.create(String)

Java adds class-relationship queries: get_sub_classes(cls), get_extended_classes(cls), get_implemented_interfaces(cls), plus get_all_crud_operations() and get_all_comments().

Call graph & reachability

Build at the right level

These methods need analysis_level=AnalysisLevel.call_graph. The default symbol_table level does not populate call edges.

Method	Returns	Notes
get_call_graph()	networkx.DiGraph	Edges point caller → callee.
get_call_graph_json()	str	Same graph, serialized.
get_callers(target_class_name, target_method_declaration)	Dict	Who calls this method.
get_callees(source_class_name, source_method_declaration)	Dict	What this method calls.
get_class_call_graph(qualified_class_name, method_signature=None)	graph	Call graph scoped to one class (Java).

cg = analysis.get_call_graph()          # -> networkx.DiGraph
callers = analysis.get_callers(
    "org.apache.commons.cli.Options", "addOption(Option)"
)  # -> Dict of call sites

Reachability is a networkx query over the call graph:

import networkx as nx

cg = analysis.get_call_graph()
print(nx.has_path(cg, src, sink))                 # reachable? -> True / False
# nx.shortest_path(cg, src, sink)                 # one path
# list(nx.all_simple_paths(cg, src, sink))        # every path

Models

The typed objects each facade returns.

Language	Models
Java	JType, JCallable, JCompilationUnit, JField, JComment, JCRUDOperation
Python	PyModule, PyClass, PyCallable, PyImport, PyComment

Gotchas

Read these once, save yourself an hour

• Call-graph level is required. get_call_graph, get_callers, get_callees, and reachability need analysis_level=AnalysisLevel.call_graph. The default symbol_table won’t have call edges.
• Analysis is project-based. Always pass analysis(project_path=...); point it at the project root.

Keep going

Common tasks Copy-paste snippets for each call above.

cocoa Build a Code Context Agent plugin on these methods.

Python API reference Every method on every analysis facade.

CLDK over MCP Expose the same analysis layer as MCP tools.