feat(mcp): expose blocks.delete/deleteRange via superdoc_edit

[jacobjove]

Summary

Wires the existing document-api blocks.delete / blocks.deleteRange ops into the MCP tool surface as two new superdoc_edit actions — delete_block and delete_block_range — so an MCP client can remove an entire block node, not just its text. Closes the gap where deleting a heading/paragraph's text leaves an empty container that still renders as block spacing.

Proposed in #3586 (surface options A/B/C discussed there). This PR implements option A.

What changed

• Contract (packages/document-api/src/contract/operation-definitions.ts): intentGroup:'edit' + intentAction:'delete_block' on blocks.delete; intentGroup:'edit' + intentAction:'delete_block_range' on blocks.deleteRange; one disambiguation sentence and one inputExample in INTENT_GROUP_META.edit.
• Regenerated via pnpm run generate:all: apps/mcp/src/generated/{catalog,intent-dispatch.generated}.ts, packages/sdk/langs/browser/src/intent-dispatch.ts, packages/sdk/tools/intent_dispatch_generated.py.
• Tests (apps/mcp/src/__tests__/protocol.test.ts): action-enum assertion plus three end-to-end cases asserting physical block removal (block-count decrement + nodeId absence) — one demonstrating the text-delete-leaves-empty-container bug vs delete_block removing it, and one asserting delete_block_range removes a contiguous top-level span (deletedCount + spanned nodeIds absent, with a neighbor outside the range surviving so the delete is proven bounded).

Addressing

delete (existing) takes a text ref; delete_block takes target:{kind:'block',nodeType,nodeId}; delete_block_range takes start/end (each a {kind:'block',nodeType,nodeId} top-level block, inclusive). The delete_block name is forced by the existing delete collision in the edit group.

Verification

apps/mcp 41 pass (4 new) · document-api 1480 pass (blocks + contract-parity) · conformance 1198 pass · codegen incl. cross-lang-parity (Node↔Python) 57 pass · typecheck 0 new errors vs baseline.

Limitations

• supportsTrackedMode: false, so block-node removal isn't recordable as a tracked change (cf. feat(track-changes): block-level structural tracked changes for tables #3343 for tables).
• Does not add a superdoc_mutations block.delete step (option C); a sound follow-up.
• blocks.delete overlaps tables.delete/lists.delete as the type-agnostic primitive; the type-specific tools remain.
• The delete_block_range contract entry carries the disambiguation sentence but not its own inputExample (the edit group example covers delete_block); deferred until the maintainer surface decision in MCP: expose block-node deletion (blocks.delete / blocks.deleteRange) via superdoc_edit #3586 (option A vs a dedicated superdoc_blocks tool vs a superdoc_mutations block.delete step) is settled, to avoid regenerating prose that a surface change would rewrite.

[caio-pizzol]

Thanks for the contribution, Jacob! We reproduced the block-deletion gap in the released MCP package. The newer core path handles selected paragraph deletion on main, but it hasn’t shipped, and we haven’t established equivalent range support. We’re keeping this PR open while we work through those remaining gaps.

[github-actions]

Thank you for your submission, we really appreciate it. Like many open-source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution. You can sign the CLA by just posting a Pull Request Comment same as the below format.

---

I have read the CLA Document and I hereby sign the CLA

---

Jacob Jove seems not to be a GitHub user. You need a GitHub account to be able to sign the CLA. If you have already a GitHub account, please add the email address used for this commit to your account.
_{You can retrigger this bot by commenting recheck in this Pull Request. Posted by the CLA Assistant Lite bot.}