ref(mcp): Separate tool, prompt and resource patches

[alexander-alderman-webb]

Description

Issues

Reminders

• Please add tests to validate your changes, and lint your code using uv run ruff.
• Add GH Issue ID & Linear ID (if applicable)
• PR title should use conventional commit style (feat:, fix:, ref:, meta:)
• For external contributors: CONTRIBUTING.md, Sentry SDK development docs, Discord community

[github-actions]

Codecov Results 📊

✅ 90048 passed | ⏭️ 6298 skipped | Total: 96346 | Pass Rate: 93.46% | Execution Time: 313m 29s

📊 Comparison with Base Branch

Metric	Change
Total Tests	—
Passed Tests	—
Failed Tests	—
Skipped Tests	—

✨ No test changes detected

All tests are passing successfully.

❌ Patch coverage is 69.51%. Project has 2477 uncovered lines.
❌ Project coverage is 89.65%. Comparing base (base) to head (head).

Files with missing lines (1)

File	Patch %	Lines
sentry_sdk/integrations/mcp.py	69.51%	⚠️ 50 Missing and 9 partials

Coverage diff

@@            Coverage Diff             @@
##          main       #PR       +/-##
==========================================
- Coverage    89.77%    89.65%    -0.12%
==========================================
  Files          192       192         —
  Lines        23824     23923       +99
  Branches      8226      8250       +24
==========================================
+ Hits         21386     21446       +60
- Misses        2438      2477       +39
- Partials      1349      1354        +5

---

Generated by Codecov Action

[sentry-warden]

handler_name may be unbound in _extract_handler_data_from_args for tool and prompt types

In _extract_handler_data_from_args (mcp.py ~line 423), if original_args is empty and original_kwargs.get("name") is falsy, handler_name is never assigned for handler_type == "tool" or "prompt", causing an UnboundLocalError when it is referenced at return handler_name, arguments (tool) or arguments = {"name": handler_name, ...} (prompt). Add an else: handler_name = "unknown" fallback in both branches.

Evidence

• _extract_handler_data_from_args lines 422-446: both tool and prompt branches assign handler_name only under if original_args: or elif original_kwargs.get("name"):, with no else clause.
• For the prompt branch, handler_name is immediately used on the very next line: arguments = {"name": handler_name, **(arguments or {})}, so an UnboundLocalError is raised before return.
• For the tool branch, handler_name is used in the shared return handler_name, arguments at line ~455.
• _prepare_handler_data reaches _extract_handler_data_from_args when both params is None and _is_v2_context(original_args) is False (e.g., empty original_args in a v2 kwargs-only call).
• The resource branch correctly initialises handler_name = "unknown" before any conditional, confirming the fix pattern is known.

_{Identified by Warden find-bugs}

[sentry-warden]

UnboundLocalError when tool/prompt handler called with no args and no kwargs name

In _extract_handler_data_from_args, handler_name is only assigned inside if original_args: / elif original_kwargs.get("name"): branches for the "tool" and "prompt" types; if both conditions are false, handler_name is never set. The prompt branch then immediately dereferences it at arguments = {"name": handler_name, ...}, raising UnboundLocalError at runtime.

Evidence

• _extract_handler_data_from_args (mcp.py ~line 407) handles "tool" and "prompt" with if original_args: handler_name = ... / elif original_kwargs.get("name"): handler_name = ... — no else clause and no default value.
• For "prompt", line 446 unconditionally reads handler_name in arguments = {"name": handler_name, **(arguments or {})} before the function returns.
• For "tool", the unbound name would surface at return handler_name, arguments.
• The "resource" branch correctly defaults to handler_name = "unknown" first, so only tool and prompt are affected.
• _prepare_handler_data calls this function on the non-v2 code path (line 486) without any guard against empty args.

_{Identified by Warden code-review}

[sentry-warden]

UnboundLocalError when original_args is empty and original_kwargs has no 'name' key in _extract_handler_data_from_args

For handler_type == 'tool' and handler_type == 'prompt', handler_name is assigned only inside if/elif branches with no else fallback; if both conditions are falsy the variable is unbound and will raise UnboundLocalError — for 'prompt' the crash occurs immediately at arguments = {'name': handler_name, ...} inside the function.

Evidence

• _extract_handler_data_from_args (mcp.py ~line 421): if original_args: handler_name = ... / elif original_kwargs.get('name'): handler_name = ... — no else clause for either the 'tool' or 'prompt' branch.
• The 'resource' branch correctly initialises handler_name = 'unknown' before the conditionals, showing the pattern is inconsistent.
• For 'prompt', handler_name is dereferenced immediately within the same function at arguments = {'name': handler_name, ...} (line ~445), so an empty call with no positional args and no kwargs 'name' crashes before returning.
• _prepare_handler_data calls this function via the else branch when params is None and _is_v2_context is False, so v1 stdio callers with missing arguments hit this path.

_{Identified by Warden find-bugs}