lsp-concurrency-audit

Requires the agent-lsp MCP server.

lsp-concurrency-audit

Given a type or file, map all fields, identify which are accessed from multiple concurrent contexts, and flag fields that lack synchronization. Produces a field-level concurrency safety report.

When to Use

• Before refactoring a type that is accessed from goroutines/threads
• Auditing a codebase for data race candidates
• Reviewing a PR that adds concurrent access to an existing type
• Understanding which fields in a type need mutex protection

Input

/lsp-concurrency-audit <file-path> [--type <TypeName>]

If --type is provided, audit only that type. Otherwise, audit all types in the file that have concurrent callers.

Step 1: Discover types and fields

Call list_symbols on the target file to enumerate all types (structs, classes):

mcp__lsp__list_symbols({ "file_path": "<target>" })

For each type (kind=23 struct, kind=5 class), collect:

• Type name
• All fields (children with kind=8 field or kind=7 variable)
• Whether any field's name or detail contains sync primitives ("Mutex", "RWMutex", "Lock", "Semaphore", "atomic", "Atomic", "sync.", "pthread_mutex", "std::mutex")

If --type was specified, filter to that type only.

Step 2: Blast radius and sync-guarded status

Call blast_radius on the file:

mcp__lsp__blast_radius({
  "changed_files": ["<target>"],
  "scope": "all"
})

From the result, for each method on each target type:

• Record sync_guarded: true/false from the response
• Record non_test_callers count (blast radius)
• Record test_callers count

Step 3: Trace concurrent boundaries

For each method on each target type, call find_callers with cross_concurrent: true:

mcp__lsp__find_callers({
  "file_path": "<target>",
  "line": <method_line>,
  "column": <method_column>,
  "direction": "incoming",
  "cross_concurrent": true
})

Record for each method:

• concurrent_callers: list of callers that cross concurrent boundaries
• pattern: the concurrent entry pattern detected (e.g., "go func(", "Thread.start(")