Productivity

Parsewise

by Parsewise

Overview

Extract data from document sets with citations. Turn collections of contracts, reports, invoices, and other documents into structured results you can trace back to the source. Define the fields you need, compare information across documents, calculate derived metrics, and investigate missing or inconsistent results. Organise your document sets in Parsewise projects and track extraction progress directly from your assistant.

Tools

archive_project

ChatGPT

Archive a live project. This hides it from list_projects and live project-scoped tools without deleting documents, agents, or results. Use restore_archived_project to make it live again.

cancel_agent_pipeline

ChatGPT

Cancel queued or in-flight agent pipeline work started by launch_agents. This bumps the project run number so stale tasks skip future work, clears queued/running pipeline flags, and releases any reserved pipeline credit estimate. It does not delete agents, documents, or completed results.

create_agent

ChatGPT

Create one agent that produces a result field for a project. An extraction agent (the default) reads the value out of the project documents and needs extraction_instructions. A derived agent instead computes its value from other agents: pass agent_type="derived" with a derived_prompt and the derived_agent_references it reads, and use it for roll-ups, comparisons, ratios, and classifications over already-extracted values rather than for reading documents. Referenced agents must already exist in the project, and reference cycles are rejected. Add dimensions when the field should produce repeated values, such as one value per document, party, clause, or region. Custom dimension settings and values are synchronized across agents sharing that dimension. Unattached agents are unaffected unless add_to_all_agents is true. Supplying a replacement value list or switching to dynamic values can delete shared dimension values and associated extraction results, even when add_to_all_agents is false. Deleted results require re-extraction; deleting the new agent does not restore them.

create_dimension

ChatGPT

Create a project dimension such as Year, Region, Party, or Clause. Use the returned id when configuring agent dimensions.

create_project

ChatGPT

Create a project to hold source documents, agents, and results. Set auto_attach_document_dimension=true when each document should become its own result row: the system-managed Document dimension then attaches automatically to every agent created in the project.

create_project_folder

ChatGPT

Create a project folder. Nest it by passing parent_id, or omit parent_id for a root folder.

delete_agent

ChatGPT

Destructively delete one agent from a live project. This removes the agent configuration and cascades through its dimension mappings, dimension instances, extraction data, and resolved results.

delete_dimension

ChatGPT

Destructively delete a project dimension. This cascades through dimension mappings and extracted dimension values: agents survive, but this dimension is detached from every agent on the project and stored values for this dimension are cleared.

delete_document

ChatGPT

Destructively delete one source document from a live project. This removes the document, its pages, storage files, document-specific tags, and extraction data derived from that document. If the deleted document was pending, remaining pending documents may be auto-started.

delete_project_folder

ChatGPT

Delete a project folder and all of its empty subfolders. The hierarchy cannot contain live projects. Archived and deleted projects do not block deletion and become unfiled.

finalize_document_upload

ChatGPT

Finalize 1 to 25 upload slots after each supplied file has been PUT to its upload_url. Only supplied IDs are finalized, atomically; omitted slots can be finalized later. Ingests files and starts parsing. A file may yield zero, one, or several response items (for example spreadsheet sheets or email attachments). Fails if no new documents are ingested.

get_agent

ChatGPT

Get one agent definition, including extraction instructions, value type, examples, web-search setting, and dimensions.

get_agent_processing_status

ChatGPT

Get per-agent processing status for a project, including parsing, extraction, web-search, resolution progress, consistency counts, and whether the agent pipeline is running.

get_dimension

ChatGPT

Get one project dimension by ID, including description, example values, and value type.

get_document_page

ChatGPT

Get extracted text from one 1-indexed document page. Use this to read document content page by page; start_char and max_chars page through unusually large extracted pages.

get_document_summary_and_metadata

ChatGPT

Get a document overview by project_id and document_id, including the generated summary, metadata, parsing status, tags, file name, file URL, and page count. This does not return full document text.

get_project_processing_status

ChatGPT

Get compact processing status for a project, including document parsing state, pending documents, agent pipeline state, and schema generation status. Use for frequent polling.

get_result

ChatGPT

Get one resolved agent result, including value, status, dimensions, citations, source document pages, and review metadata. result_id is the resolution_result.id field on a list_results row (not the row's atomic_value_id).

get_usage_limits

ChatGPT

Get usage and quota limits for a project. Returns project-scoped document, agent, and pipeline-size limits plus organization-level page or credit context used before launching agents. This may prepare internal extraction cardinality records, so it is not advertised as read-only-safe.

launch_agents

ChatGPT

Launch the agent extraction pipeline for a project after creating or updating agents. Returns immediately; poll agent status or results to track progress. A project with failed documents cannot launch until those files are removed or corrected replacements are uploaded.

list_agents

ChatGPT

List agents in a project. Agents define what to extract from documents and how to normalize it. Returns compact summaries; use get_agent for full instructions, examples, and dimension values.

list_archived_projects

ChatGPT

List archived projects in the organization. Archived projects are hidden from list_projects and most project-scoped tools until restored with restore_archived_project.

list_dimensions

ChatGPT

List project dimensions. Create or inspect dimensions before passing dimension_id values to create_agent or update_agent.

list_documents

ChatGPT

List source documents in a project, most recent first. Use this to find document IDs, page counts, parsing status, tags, and file metadata; use get_document_page to read extracted page text.

list_project_folders

ChatGPT

List project folders in the organisation as a flat list. Assemble the tree from parent_id. Use folder IDs with create_project, update_project, and create_project_folder.

list_projects

ChatGPT

List projects in the organization. Start here to find project IDs for document, agent, and result tools. Returns project summaries with document counts, agent counts, and whether results exist.

list_result_sources

ChatGPT

List the source passages, reasoning, consistency checks, and citations that contributed to one resolved result. result_id is the resolution_result.id field on a list_results row.

list_results

ChatGPT

List resolved agent results for a project. Returns compact result rows across one or more agents. Use get_result for one resolved value and list_result_sources for the source passages behind it.

request_document_upload

ChatGPT

Start uploading one or more local documents to a project (up to 25 files per call). Returns a short-lived upload_url per file; PUT each file to its upload_url (curl --data-binary @path), then call finalize_document_upload with the uploaded IDs to finalize together. Single file: file_names=["doc.pdf"]. Supported: pdf, doc(x), ppt(x), xlsx, png, jpg/jpeg, eml, txt, md/markdown (UTF-8). File names must be unique within the batch (duplicates fail here); names that already exist in the project are accepted here but skipped at finalize (finalize fails if nothing new is ingested). Max 250MB per file.

restore_archived_project

ChatGPT

Restore an archived project to live status so it appears in list_projects and can be used by project-scoped MCP tools again.

update_agent

ChatGPT

Update an agent definition and optionally replace its dimensions. Changing instructions, derived prompt or references, value type, examples, unit, web search, complex calculations, or dimensions can delete existing extraction results; run extraction again to regenerate them. Custom dimension changes can also delete values and results for other agents sharing the dimension. Restoring the old definition does not restore deleted results. Pass an empty string to clear optional text fields such as unit, examples, inconsistency_instructions, or resolution_instructions. Converting an agent between extraction and derived is allowed: set agent_type along with the fields the new type needs.

update_dimension

ChatGPT

Update a project dimension. Omitted fields stay unchanged; renaming uses the same uniqueness and reserved-name rules as create_dimension. Overwrites shared dimension metadata used by every attached agent; previous field values are not retained as an undo history. Renaming can break consumers that refer to the old label. Existing extracted values are preserved, and relaunching alone may reuse completed results. For a type change that must regenerate results, make the change through update_agent dimensions instead of this tool, then launch_agents. Re-submitting an already saved type does not invalidate existing results.