document

This tool covers document CRUD, tree navigation, metadata, and daily-note oriented document operations.

When to read this page: you need to create, move, query, or convert documents.

Related pages:

• Path Semantics
• Permissions

Common Actions

Group	Actions
Create and read	create, lookup, ensure_link_targets, get_doc, get_outline
Tree navigation	get_child_blocks, get_child_docs, list_tree, search_docs
Metadata and mutations	rename, move, reorder, remove, set_attr, duplicate
Daily note / conversion	create_daily_note, heading_to_doc, doc_to_heading

Parameters and Semantics

• create takes either a human-readable path, or parentPath + title; omit markdown to create an empty document. Prefer path for child documents. The parentPath + title mode accepts either a human-readable parent path or a storage path ending in .sy returned by lookup.
• lookup resolves by id, storage path, or human-readable hpath / hPath; use include to request id, ids, path, hpath, or docInfo.
• ensure_link_targets provisions a reusable import link map inside one exact scope: notebook + direct-parent parentId. resolve and reuse accept only explicit direct-child document IDs. create accepts explicit new titles, but never treats a matching existing title as identity: it reports same_title_child_requires_explicit_id in unresolved instead. Every resolved or created output includes exact id, notebook, storage path, and hPath readback.
• ensure_link_targets(dryRun=true) is valid only for mode="create"; it inspects and reports wouldCreate without mutation. resolve and reuse are read-only discovery operations. A real create follows the normal two-call contract: preflight with validateOnly=true, then exactly one request with a fresh UUIDv7 requestId and returned expectedStructureHash. Do not retry after an unknown transport outcome: inspect or reuse the original request ID. The action never automatically removes documents it created.
• The returned idPath includes available id / ids. When several documents share the same hpath, include: ["ids"] returns all matching IDs; the tool includes a SQL fallback.
• rename, remove, and move often need a storage path if you are not using document IDs.
• reorder takes a notebook or parent-document parentID plus orderedIDs. The array must contain every visible direct child document ID exactly once. It enables custom notebook sorting (sortMode: 6) and does not move, rename, or edit any document.
• get_child_docs requires a document id; it does not accept notebook + path.
• list_tree uses notebook + path, and path is a storage path such as / or /20240318112233-abc123.sy, not a human-readable path.
• If bulk remove hits SiYuan's short indexing window, retry by deleting one document at a time with notebook + storage path.
• set_attr writes document metadata attributes by document ID.
• get_outline calls SiYuan's native outline endpoint and returns the heading tree, block IDs, nesting, and headingCount without reading the document body. Use get_doc instead when you also need editable Markdown.

Markdown and Title Rules

• create markdown does not need a same-name leading # Title; if present, it is stripped to avoid duplicate visible titles.
• create accepts ((id 'title')), naked ((id)), and #tag# directly. Naked refs are expanded to explicit anchors; if lookup fails, MCP falls back to ((id 'id')) with a warning.
• create allows footnote-style refs such as [^1] and [text](siyuan://blocks/id), but the result includes a hint because they do not create SiYuan backlinks.
• get_doc returns the same editable Markdown shape as fs.read, preserving ((id 'title')) and #tag#.
• get_doc mode="markdown" always returns complete display-block windows. Continue with nextWindow or use blockStart, blockLimit, and tokenBudget; the response also includes a full heading outline. includeBlockIds=true adds sidecar block references without changing content.
• Character page/pageSize pagination has been removed. mode="html" remains the unpaginated current-view HTML path and continues to use size.

Safety Rules

• remove and move require explicit confirmation.
• Always resolve document path type before mutating by path.
• Never use a title search result as a link target. Re-run ensure_link_targets in resolve or reuse mode with the exact ID returned by an earlier successful result.

Examples

MCP:

{
  "action": "create",
  "notebook": "<notebook-id>",
  "path": "/Inbox/Weekly Note",
  "markdown": "Weekly report body"
}

{
  "action": "lookup",
  "id": "<doc-id>",
  "include": "path"
}

{
  "action": "ensure_link_targets",
  "notebook": "<notebook-id>",
  "parentId": "<parent-document-id>",
  "mode": "reuse",
  "targets": [{ "key": "source-index", "id": "<direct-child-document-id>" }]
}

For a new target, call the same action first with mode: "create" and validateOnly: true, then commit exactly once using the returned expectedStructureHash, a fresh UUIDv7 requestId, and an explicit title target. A same-title child is an unresolved result, not an implicit reuse.

CLI:

siyuan document create --notebook <notebook-id> --path "/Inbox/Weekly Note" --markdown "Weekly report body"
siyuan document lookup --id <doc-id> --include path
siyuan document reorder --parent-id <notebook-or-parent-doc-id> --ordered-ids-json '["<doc-id-1>","<doc-id-2>"]'
siyuan document ensure_link_targets --notebook <notebook-id> --parent-id <parent-document-id> --mode reuse --targets-json '[{"key":"source-index","id":"<direct-child-document-id>"}]'

Action List

• create
• lookup
• ensure_link_targets
• rename
• remove
• move
• reorder
• get_child_blocks
• get_child_docs
• set_attr
• list_tree
• search_docs
• get_doc
• get_outline
• create_daily_note
• duplicate
• heading_to_doc
• doc_to_heading