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

[alexander-alderman-webb]

Description

Separate patches for tool, prompt and resource handlers.
This reduces conditionals and removes inappropriate generalization. Each handler type has different input and output types, and require different span attributes.

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 📊

✅ 90045 passed | ⏭️ 6302 skipped | Total: 96347 | Pass Rate: 93.46% | Execution Time: 316m 18s

📊 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 68.29%. Project has 2478 uncovered lines.
❌ Project coverage is 89.64%. Comparing base (base) to head (head).

Files with missing lines (1)

File	Patch %	Lines
sentry_sdk/integrations/mcp.py	68.29%	⚠️ 52 Missing and 9 partials

Coverage diff

@@            Coverage Diff             @@
##          main       #PR       +/-##
==========================================
- Coverage    89.74%    89.64%     -0.1%
==========================================
  Files          192       192         —
  Lines        23823     23911       +88
  Branches      8226      8244       +18
==========================================
+ Hits         21378     21433       +55
- Misses        2445      2478       +33
- 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}

[cursor]

Cursor Bugbot has reviewed your changes and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit 93ea189. Configure here.}

[alexander-alderman-webb]

this keeps changing when I commit, I think it's the pre-commit.