How to connect Autosheet MCP with DeepSeek Harness

Start an Autosheet agent specialized in Google Sheets work that operates on a Google Sheets spreadsheet based on a natural-language prompt. Use this tool for work inside a Google Sheets spreadsheet: reading data and inspecting its structure, everyday spreadsheet work, web research, and row-by-row bulk processing with AI. The agent reports a summary of its work rather than raw cell data. The agent is autonomous. The agent explores the spreadsheet: it locates and inspects the relevant tabs and ranges. You do not need to compute ranges, row numbers, or per-cell values for it. It plans and performs read and write operations as needed, and verifies the results. To get a plan or a preview of a write operation, request it explicitly. If the task is very large or ambitious, it is better to break it down in several follow-ups. Runs take from a few seconds to several minutes, and results may vary between runs on the same instruction. The call waits up to 60 seconds for the agent to finish. Short runs complete inside that window and return a summary of work performed along with an agent_id, so no further call is needed. Longer runs return early with the agent_id and a progress snapshot; continue with autosheet_get_agent until the run is complete. Each call starts a fresh agent with no memory of previous runs or of this conversation. If an agent from this conversation already exists and the task builds on its work, use autosheet_follow_up_agent instead. The agent handles everyday spreadsheet work: creating, fixing, and explaining formulas; analyzing data and detecting errors; building charts, pivot tables, and dashboards; looking up data; merging, moving, copy/pasting, and restructuring sheets, columns, rows, and cells; cleaning and deduplicating data; and applying formatting, validation, and conditional formatting. The agent also handles row-by-row bulk processing with AI models pre-configured by the user across large datasets: enrichment through web research, categorization, translation, scoring, and content generation or rewriting. You do not need to supply the values. It usually knows when to use AI processing vs deterministic formulas, but you can explicitly request either. The agent can also do web search similarly to non spreadsheet-based agents and cite sources. Do not confuse it with the bulk row-by-row web search feature. Read-only and bulk processing operations are safe and cannot overwrite existing data. For non-bulk write operations the agent may add, overwrite, move, or clear spreadsheet content. Earlier versions can be restored from the spreadsheet's version history. The agent will usually recover from its own mistakes but, in rare cases, if it gets stuck, you might have to steer it in the proper direction. The agent will usually not return range content to you because that would be very token inefficient. If you want to check some specific values or formulas, ask for them explicitly, sparingly. The agent cannot currently use existing cell colors as input criteria; scrape specific URLs; analyze images; display sources for bulk web search; record or run macros; run Apps Script or other add-ons; export, download, convert spreadsheet files; or create new files.

Send a follow-up prompt to an existing Autosheet agent, identified by its agent_id from autosheet_start_agent_google_sheets_spreadsheet. The agent retains full memory of its previous work in this session: what it explored, what it changed, and why. This applies whenever the request builds on, corrects, refines, or extends work the agent already did ("now also do X", "you used the wrong column", "apply the same to the other tab"). Start a new agent instead when the task is unrelated to the previous run, targets a different spreadsheet, or when the previous agent failed in a way that suggests its understanding of the sheet is wrong. Like autosheet_start_agent_google_sheets_spreadsheet, this resumes an autonomous agent: the agent takes time, may act on the spreadsheet immediately, and biases to action. You don't need to repeat context the agent already has, but be explicit about anything new. The call waits up to 60 seconds for the agent to finish. Short runs complete inside that window and return a summary of work performed along with an agent_id, so no further call is needed. Longer runs return early with the agent_id and a progress snapshot; continue with autosheet_get_agent until the run is complete. If the agent has expired, this returns an error — start a new agent.

Fetch the current status and last turn messages for an Autosheet agent. Status can be available (it has finished its latest turn and can be followed up) or busy (it is still running and cannot yet receive a follow-up).

Stop a running Autosheet agent. Use when the agent is clearly going the wrong path or the user asked to stop it explicitly.

Copy a single tab (sheet) of a Google Sheets spreadsheet into a destination spreadsheet, and return the new tab's sheet_id and title. The destination may be the same spreadsheet — that duplicates the tab in place — or a different one. Requires write access to the destination (and read access to the source when they differ). IMPORTANT: Exactly one of the sheet_id or sheet_name parameters is required to call this tool. If the source spreadsheet URL contains a gid parameter, use it to set the sheet_id. Never use a gid from the destination spreadsheet URL. If you do not know and cannot guess the sheet_id or sheet_name from the context, list available tabs first with an autosheet agent before calling this tool. The copied tab keeps its values, formulas, formatting, notes, and embedded charts, and is appended as the last tab of the destination. References can break in the destination if other tabs are missing, that is expected. The tab never merges into or overwrites an existing tab. Each call adds a new tab, so verify before retrying. This is a direct, deterministic and fast operation.