Tools reference

Docsumo MCP exposes 100+ tools. Your agent picks the right one from your prompt — you never call these directly, but knowing what's available helps you phrase requests. This page covers the main ones, so you know what to expect. All names are prefixed with docsumo_.

IDs flow between tools: get_document_types returns the doc_type values uploads and filters take, list_documents returns doc_id, list_casetypes returns casetype_id, list_cases returns case_id, and list_workflows returns workflow_id.

Tools act as the account behind your API key, so your role and plan still apply; downloads return a secure, time-limited link rather than raw file bytes; long lists come back a page at a time; and uploads and reprocessing spend document credits just like the web app.

Documents

Tool	What it does
`upload_document`	Upload one file for a given `doc_type`. Optional `folder_id`/`folder_name`, `password` for protected PDFs, `user_doc_id` and `doc_meta_data` for your own tracking, `mode` (`prod`/`test`). Extraction runs asynchronously — the agent comes back when it's done.
`upload_case_document`	Upload a packet of up to 30 files to a `casetype_id` — the way to ingest real cases. Creates a case, or adds to one with `case_id`. Optional `case_name`, `case_fields`, and `trigger_workflow` (on by default).
`list_documents`	Find documents. Filter by `doc_type`, `status` (`new`, `processing`, `review_required`, `reviewing`, `review_skipped`, `processed`, `erred`), `created_date_gte`/`_lte`, `folder_id`, uploader, with paging and `sort_by`.
`get_document_data`	The main result tool — extracted key-value fields and line-item tables for one `doc_id`, including human corrections.
`download_document`	Up to 25 `doc_ids` in one call. `format` is the file itself (`original_file`, `searchable_pdf`, `images`, `ocr_text`) or the extracted data (`json`, `csv`, `xml`). One document failing doesn't sink the rest.
`download_case_documents`	The same, for every document in a `case_id` — no need to know the doc IDs. Pass `doc_ids` to narrow it.
`doctype_retry_documents`	Reprocess documents, running extraction again. `trigger_workflow` re-runs the document's case workflow instead.

"Upload these invoices, then give me the line items once they're done."

Documents can also be moved between folders, assigned for review, and switched to another document type.

Document types

Tool	What it does
`get_document_types`	Every type on the account with document counts. `q` filters by title or identifier.
`get_doctype_schema_and_fields`	Read a type's sections, fields, prompts and dropdowns. `summary: true` returns names, ids and types only — much smaller.
`add_doctype_fields` / `update_doctype_fields`	Add or change extraction fields, keyed by section name — including the prompt behind each field.
`configure_update_setting`	Update any Configure setting (General, Pre-processing, Extraction, Post-processing, Import & Export) by `setting_name` and `value`.
`field_setup_run_test`	Try a field's configuration against a real document before saving it.
`get_doctype_analytics`	Usage, processing time and field-level accuracy. `report_type`: `daily` (last 7 days), `weekly` (last 7 weeks), or `range` with `from_date`/`to_date`.

"Add a Vendor Tax ID field to the Invoice type and test it on this sample."

New types can also be added from the AI Model Hub catalog, or created blank as a custom type.

Cases

Tool	What it does
`list_cases`	Cases for a `casetype_id`, with search and paging. Filter by `stage`, `assigned_to`, `workflow_state`, or created/modified dates.
`create_case`	Create an empty case under a case type, optionally named.
`update_case`	Change `case_name`, `stage`, or `assigned_to` for named `case_ids` — or every case in a case type with `select_all`, minus any `deselected_case_ids`.
`get_case_overview`	Where one case stands: its stage, and how its documents are progressing.
`list_case_documents`	Everything attached to a case, each with its status and review URL.
`get_case_validations`	Run the case's validation checks and return what they found. `result_in` narrows to e.g. `["Review", "Fail"]`.

"Create a case for this loan application and attach these documents."

Workflows

Tool	What it does
`list_workflows`	Find a workflow. Filter by `status` (`draft`, `published`, `archived`), linked `casetype_id`, or run dates.
`get_workflow`	Read how a workflow is built. `summary: true` gives metadata, publishability and a block list instead of the full graph.
`force_run_case`	Run a case's workflow fresh, cancelling anything in flight. `workflow_block_id` starts partway through.
`rerun_case`	A cheaper re-run that reuses the stored extraction instead of processing documents again.
`list_workflow_executions`	Search runs, newest first. Filter by workflow, `status` (`processing`/`success`/`error`), `trigger` (`upload`, `api`, `email`, `manual`, `schedule`, `case_event`), or `mode`.
`get_workflow_execution_history`	Block-by-block history for one `execution_id` — where a run went and where it stopped.
`update_workflow_approvals`	Approve or reject a case parked at a Human-in-the-Loop step: `block_id`, `approved`, optional `case_id` and `reason`.

"Run the onboarding workflow on this case and tell me where it stops."

The server also exposes the tools the Docsumo workflow builder uses — editing a graph, generating block code, managing test files. Agents use those while building a workflow; you don't call them directly.

Tables and lookups

Tool	What it does
`list_tables`	Your tables, by `kind`: `reference` (lookup data like GL codes or currencies) or `entity` (records such as vendors). `q` searches table names, not rows.
`get_table_rows`	A page of one table's rows. `q` is a typo-tolerant search across every cell; `conditions` + `match` filter by column; `resolve_links` shows linked records by name rather than row id.
`search_tables`	One search across every table at once, grouped by table.
`list_database_tables` / `get_database_table_rows`	The CSV lookup tables behind dropdown field options, and a searchable page of one table's rows.
`append_database_table_rows`	Add rows to a Database Table. A cell naming a column the table doesn't have is dropped silently, so check the columns first.
`update_database_table_cell`	Set one cell: `row_id`, `column`, `value`.

Tables are read-only through MCP — rows are authored on the Tables screen and written by workflows. Database Tables can be edited; deleting rows or renaming a table requires an admin role.

Account

Tool	What it does
`get_user_detail`	Your identity, plan, document allowance and usage. Good first call to verify the connection.
`list_account_members`	Everyone in the account with their roles and document type access; `query_fullname` finds one person.
`get_webhook_config` / `update_webhook_url`	Read or change the organization's webhook endpoint and its authentication.
`list_activity_logs`	The All Activity log by `activity_type` (`all`, `document`, `user`, `credit`, `webhook`), with text search, date range, actor and action filters.

API keys are not managed through MCP — create and rotate them in Settings → API & Webhooks in the web app.