Skip to main content

​
init_plan

Initialize revision-bound context metadata for a document, enabling coordinated multi-agent planning. The returned plan object captures the current document revision so that downstream tools can detect stale edits. Hints: readOnly — does not modify the document.
​
file_path
string
required
Path to the DOCX file.
​
plan_name
string
Human-readable name for the plan. Useful for logging and debugging in multi-agent workflows.
​
orchestrator_id
string
Identifier for the orchestrating agent. Stored in the plan metadata.

​
merge_plans

Deterministically merge multiple sub-agent plans into a single ordered plan and detect hard conflicts before any edits are applied. Use this before calling apply_plan when multiple agents have independently produced edit plans against the same document. Hints: readOnly — does not modify any documents.
A hard conflict occurs when two plans attempt to edit the same paragraph in an incompatible way, for example both replacing different text in the same _bk_* node. Conflicts are reported before apply so you can resolve them without touching the document.
​
plans
object[]
required
Array of plan objects returned by init_plan or produced by sub-agents. Each object must be a valid plan structure.
​
fail_on_conflict
boolean
When true, return an error immediately if any hard conflict is detected instead of returning a partial merged plan. Defaults to false.
​
require_shared_base_revision
boolean
When true, return an error if the plans were not all initialized from the same document revision. Helps prevent merging stale plans.

​
apply_plan

Validate and apply a batch of edit steps to a document in a single call. All steps are validated first; the tool applies them only if every step passes validation. Accepts steps inline or from a JSON file, and is compatible with the output of merge_plans.
This tool modifies the document in memory. Call save to persist the result to disk. If any step fails validation, no edits are applied.
​
file_path
string
required
Path to the DOCX file.
​
steps
object[]
Inline array of edit steps. Each step requires a step_id, an operation, and operation-specific fields. Mutually exclusive with plan_file_path.
​
plan_file_path
string
Path to a .json file containing an array of edit steps. Mutually exclusive with steps.

​
Step structure

Each step is a JSON object with a step_id, an operation (currently replace_text or insert_paragraph), and the parameters required by that operation. 
[
  {
    "step_id": "step-1",
    "operation": "replace_text",
    "target_paragraph_id": "_bk_abc123",
    "old_string": "original text",
    "new_string": "replacement text",
    "instruction": "Update the term definition"
  }
]

​
close_file

Close an open document session and release associated resources. Pass clear_all: true with confirm: true to close all open sessions at once.
Closing a session discards any unsaved in-memory edits. Call save before closing if you need to keep your changes.
​
file_path
string
Path to the DOCX file to close. Provide either file_path or google_doc_id.
​
google_doc_id
string
Google Doc ID or URL. Extract the ID from the URL: docs.google.com/document/d/{ID}/edit.
​
clear_all
boolean
When true, closes all open sessions. Requires confirm: true.
​
confirm
boolean
Required confirmation flag when using clear_all. Set to true to proceed.

Build docs developers (and LLMs) love

Get started for freeTalk to us