JLens MCP Expert Navigation Guide (V1.1.1)

JLens is a Model Context Protocol (MCP) server designed for AI Agents. It provides deep understanding of Java projects through real reflection analysis and bytecode parsing. Unlike simple text search, JLens can identify class inheritance, method signatures, visibility modifiers, and complex Maven dependency topologies.

1. Core Tools and Interaction Details

1.1 Class Inspection (inspectjavaclass)

Interaction Logic:

• State Machine Responses:

- status: "SUCCESS": Parsing successful. Read decompiledSource for logic, and methods/fields for structure. - status: "LOCALSOURCE": Important! The target class belongs to the current workspace. Ignore metadata and directly read the source code using the readfile tool via the provided sourceFile path. - status: "NOTFOUND": Class does not exist. Check the suggestion field; you might need to run buildmodule first.

• Key Parameters:

- bypassCache: If class structure seems outdated (e.g., after recent source changes), set to true to bypass GAV cache for real-time parsing.

1.2 Paginated Search (searchjavaclass)

Pagination Protocol:

• Result Parsing: If hasMore is true in the response, you must record the nextCursor.
• Subsequent Requests: Pass cursor: "<nextCursor>" in the next call until hasMore is false. Results are sorted by fully qualified name for stability.

1.3 Dependency Analysis (listmoduledependencies)

Metadata Parsing:

• Returns a standard Maven dependency model. Focus on the scope field (e.g., test, provided) to determine if a class is available at runtime.

2. Task-Driven Workflows

Scenario A: Troubleshooting NoClassDefFoundError or NoSuchMethodError

1. Locate Dependencies: Call listmoduledependencies to check if multiple versions of Jars contain conflicting classes.
2. Version-Isolated Inspection: Use JLens's multi-version isolation feature to load the class for different module contexts respectively. Compare methods lists to find missing methods.
3. Build Verification: After modifying pom.xml, call build_module to refresh the local repository and verify the build passes.

Scenario B: Massive Code Refactoring Support

1. Global Search: Use searchjavaclass with wildcards (e.g., com.api.*Service) to find all relevant interfaces.
2. Batch Inspection: Use listclassfields to quickly extract private fields of all implementation classes to evaluate state storage.
3. Source Navigation: When LOCALSOURCE is detected, automatically switch to readfile to modify code.

3. Execution Strategy

3.1 Protocol Priority

1. Probe Registered Service: First check if jlens-mcp-server exists in the environment.
2. Direct Request: If present, use standard MCP call_tool requests.

3.2 Command-Line Fallback (When no registered service)

If no service is found, the Agent must autonomously use shell tools to execute:

NPM Mode (Preferred):

npx -y @bhxch/jlens-mcp-server --tool <tool_name> --args '<json_arguments>'

Python Mode:

uvx jlens-mcp-server --tool <tool_name> --args '<json_arguments>'

4. Performance and Limitations

• GAV Cache: JLens shares cache by GroupId:ArtifactId:Version. Parsing results for third-party libraries (like Spring) are shared across projects, usually responding within 100ms.
• Initial Indexing: The first run of searchjavaclass on large projects may trigger full indexing (~60s). Subsequent requests are millisecond-level.
• JDK Requirement: Underlying environment must be Java 25+.