# Google Docs MCP [Health: Active]

**Category:** 🏢 Workplace & Productivity  
**Repository:** https://github.com/A1-x-Tech/mcp-google-docs  
**GitHub Stars:** 0  
**npm Downloads (last month):** 1086  
**Views:** 0  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/google-docs-mcp

## Description
MCP server for the Google Docs API: read and edit documents, styles, tables, images, comments.

## Tools
Capabilities this server exposes over MCP:

- **auth_status** — Shows whether this server is connected to Google: token presence and source (env variables or a stored in-chat login), expiry, the Google account email, granted vs missing OAuth scopes, where the credentials file lives and where the OAuth client comes from. Makes no network calls and never returns the token itself. Call it first when other tools report the server is not connected.
- **setup_instructions** — Step-by-step checklist for connecting this server to Google: creating a Google Cloud project and a Desktop-app OAuth client, publishing the consent screen (mandatory — Testing-mode refresh tokens die after 7 days), downloading the client JSON and handing its PATH to set_client. Works without any credentials; the checklist shortens to 'enable the API + log in' when an OAuth client is already configured (one client serves the whole mcp-google-* line). Never asks the user to paste secrets into the chat.
- **set_client** — Saves the OAuth client credentials from the JSON file downloaded from Google Cloud Console ('Download JSON' on a Desktop-app client). Pass the file PATH — the secret must never be pasted into the chat. The client is stored once in the shared ~/.config/mcp-google-auth/client.json (owner-only) and reused by every mcp-google-* server; tokens stay per-server. After this, call start_login.
- **start_login** — First step of connecting from the chat, without editing config files or restarting the client. Returns authorizeUrl — show it to the user as a clickable link and ask them to open it in the browser ON THIS MACHINE, pick the Google account and approve access. A one-shot listener on 127.0.0.1 catches Google's redirect; the code is exchanged locally and never passes through the chat. Does not open the browser itself. The attempt lives 10 minutes; when the browser shows the success page, call finish_login.
- **finish_login** — Second step: waits up to 20 s for the browser consent to finish, saves the tokens to an owner-only file and verifies the login with a read-only identity call, returning the account email and the granted scopes. After success every tool works immediately — no client restart. If the user granted only part of the requested permissions, the login is still saved and missingScopes lists what will not work. Logging in under a different Google account replaces the previous login (its refresh token is revoked best-effort) and the response carries previousAccountEmail so the change never goes unnoticed.
- **logout** — Revokes the stored token at Google (oauth2.googleapis.com/revoke) and deletes the local credentials file. Tokens supplied via env variables are NOT touched — remove them from the MCP client config manually; envTokenStillSet in the response says whether any are still in effect.
- **create_document** — Creates a new Google Doc and returns its documentId. Without markdown it calls documents.create, which accepts ONLY a title — add content afterwards with insert_text/import_markdown. With markdown, the document is created from that Markdown via Drive conversion (headings, bold/italic, links, lists, tables and code blocks become native Docs formatting) and the result carries id (the documentId), name and mimeType. The document lands in the authorized user's My Drive root; moving or sharing it needs the Drive UI or API, which this server does not cover.
- **get_document** — Returns the raw Docs API document: title, documentId, revisionId, per-tab body with every structural element's startIndex/endIndex, textRun styles, tables, lists, inlineObjects (image ids for replace_image), headers/footers (their segment ids) and named styles. This is the exact index map that range tools consume, but it is VERBOSE — for reading content prefer read_document_text, which returns compact blocks with the same indexes. include_tabs_content=true (default) populates all tabs; suggestions_view_mode controls how unresolved suggestions render (default DEFAULT_FOR_CURRENT_ACCESS).
- **read_document_text** — Returns the document as compact readable blocks, per tab: each paragraph with its text, start/end indexes (UTF-16 code units — the coordinates insert_text/replace_range/delete_range/style tools take), named style (headings) and bullet flag; tables come as cells[row][column] plain text with the table's range. Inline images appear as [image:<objectId>] placeholders (the id feeds replace_image). tab_id limits the output to one tab. Call this before any range edit — every insert/delete shifts later indexes, so indexes from before a mutation are stale.
- **list_tabs** — Returns the document's tab tree without content: tabId, title, position index and nested childTabs. Use the tabId values to target a specific tab in read_document_text and every editing tool. NOTE: the Docs API cannot create, rename, delete or reorder tabs — that is UI-only; this tool only discovers what exists.
- **export_document** — Exports the document via Drive. Text formats (markdown, txt, html) are returned inline as the tool result — format=markdown is the way to read a doc as Markdown for a round trip with import_markdown. Binary formats (pdf, docx, odt, rtf, epub) require output_path and are written to that local file (the result reports saved_to and bytes); output_path also works for text formats. output_path must be absolute, and an existing file is never overwritten unless overwrite=true. Exports are capped at 10 MB by the Drive API; comments and suggestions are not part of any export. Nothing in Google Docs changes; the only side effect is the local file when output_path is set.
- **import_markdown** — REPLACES the ENTIRE document content with the given Markdown via Drive conversion — the Markdown round-trip writer paired with export_document format=markdown. Supported Markdown (headings, bold/italic/strikethrough, links, ordered/unordered lists, tables, code blocks, images by URL) becomes native Docs formatting. Everything not expressed in the Markdown is LOST: existing text, comment anchors (comments survive but detach), positioned objects, headers/footers and extra tabs. For surgical edits use replace_range/style tools instead. Returns the Drive file resource (id, name, modifiedTime).
- **insert_text** — Inserts plain text at an index (UTF-16 code units; body content starts at index 1 — get indexes from read_document_text). Omit index to append at the end of the body (or of the segment when segment_id is set). Newlines in the text create new paragraphs; the text inherits the style at the insertion point — style it afterwards with style_text/style_paragraph. Every insert shifts all later indexes, so when making several edits, apply them from the END of the document backwards or re-read between edits. Returns the batchUpdate reply with the new revisionId.
- **replace_range** — Replaces the content in [start_index, end_index) with new text — one atomic batchUpdate that deletes the range and inserts at its start, so nothing is lost if either step would fail. Get fresh indexes from read_document_text first: indexes are stale after ANY previous edit. The new text takes the style at start_index. Cannot cut across a table cell boundary or delete a paragraph's final newline together with only part of the next paragraph — the API rejects such ranges. An empty replacement is allowed and equals delete_range.
- **replace_all_text** — Replaces EVERY occurrence of a literal string (no regex) across the whole document, or only in the tabs named by tab_ids. match_case defaults to true. Returns occurrencesChanged in the batchUpdate reply — 0 means the text was not found, not an error. Safer than replace_range for textual substitutions because it does not depend on indexes; use replace_range when position matters (e.g. only one of several occurrences).
- **delete_range** — Deletes the content in [start_index, end_index) — text, inline images, whole tables or any mix, as long as the range does not cut across a table cell boundary. Get fresh indexes from read_document_text first; every delete shifts all later indexes down. Deleting a paragraph's trailing newline merges it with the next paragraph. The tab's final newline cannot be deleted. This is final — there is no undo through the API.
- **style_text** — Applies character formatting to [start_index, end_index): bold, italic, underline, strikethrough, small_caps, font_size (points), font_family (e.g. "Roboto"), foreground_color/background_color (6-digit hex), link_url (empty string removes an existing link) and baseline_offset (superscript/subscript/none). Only the provided fields change (the fields mask is computed); at least one is required. Explicit false turns a toggle off. Get indexes from read_document_text. Styling does not shift indexes, so several style calls can safely reuse the same coordinates.
- **style_paragraph** — Applies paragraph formatting to every paragraph overlapping [start_index, end_index): named_style (normal_text, title, subtitle, heading_1..heading_6 — the way to make headings), alignment (start/center/end/justified), line_spacing (100 = single, 200 = double), space_above/space_below (points), indent_start/indent_end/indent_first_line (points), keep_with_next and direction (ltr/rtl). Only the provided fields change; at least one is required. A range inside a single paragraph styles that whole paragraph.
- **set_paragraph_bullets** — Turns the paragraphs overlapping [start_index, end_index) into a list, or removes their bullets with remove=true. Presets: disc, arrow, checkbox, star, diamond (bulleted); decimal, decimal_parens, decimal_nested, upper_alpha, upper_roman, zero_decimal (numbered). Nesting follows each paragraph's leading tabs — indent with \t in insert_text before applying bullets to create sub-items. Exactly one of preset or remove is required.
- **insert_table** — Inserts an empty rows x columns table at an index, or at the end of the body when index is omitted. Fill the cells afterwards: read_document_text shows the new table's range and per-cell layout, then insert_text into each cell (cell content ranges are visible in get_document; each cell holds its own paragraphs). Cell merging, cell background/borders and pinned header rows are not covered by typed tools — use raw_request with mergeTableCells/updateTableCellStyle/pinTableHeaderRows.
- **edit_table** — Inserts or deletes a row/column of an existing table. The table is addressed by table_start_index — the table block's start index from read_document_text (its startIndex in get_document). action=insert_row adds next to the cell at row_index/column_index (insert_below, default true); insert_column likewise (insert_right, default true); delete_row / delete_column remove the row/column containing that cell WITH ALL ITS CONTENT — that content is unrecoverable. row_index/column_index are 0-based and default to 0. Deleting shifts every later index in the document; re-read before further edits.
- **insert_break** — Inserts a break at an index (or at the end of the body when index is omitted). kind=page starts a new page; kind=section_next_page starts a new section on a new page; kind=section_continuous starts a new section on the same page. Sections carry their own margins/columns — style them via raw_request updateSectionStyle. Breaks cannot be inserted into headers, footers, footnotes or table cells. A break occupies one index position; remove one by deleting its range with delete_range.
- **insert_image** — Inserts an inline image fetched from a PUBLIC URL at an index (or at the end of the body when index is omitted). The URL must be reachable by Google without auth, at most 50 MB and 25 megapixels, in PNG, JPEG or GIF; there is no upload channel — a local file must be hosted somewhere public first. width_pt/height_pt set the displayed size in points (omit both for natural size capped to the page width; when only one is given the API scales the other to keep the aspect ratio). Returns the created objectId in the reply — keep it for replace_image. Delete an image by deleting its range with delete_range (it occupies one index).
- **replace_image** — Replaces an existing image's contents with a new image from a public URL, keeping the original's size and position (the new image is center-cropped to fit). image_object_id is the inline object id — read_document_text shows it inside [image:<objectId>] placeholders, and get_document lists all ids under inlineObjects. The same URL rules as insert_image apply (public, <=50 MB, <=25 MP, PNG/JPEG/GIF).
- **list_comments** — Lists the document's comment threads via the Drive API: for each comment its id, content, author displayName, createdTime/modifiedTime, resolved flag, the quoted document text it anchors to (quotedFileContent) and its replies (with their action: resolve/reopen). Paginate with page_token; include_deleted=true also returns deleted comments as tombstones. Requires a Drive scope on the OAuth token (comments are Drive data, not Docs API data). Comment ids feed manage_comment.
- **manage_comment** — Manages comment threads via the Drive API. action=create adds a NEW comment (needs content; optional quoted_text cites a passage — the API cannot anchor a new comment to a live range, that anchor format is not public, so it appears at document level). action=reply adds a reply (needs comment_id + content). action=resolve closes the thread (needs comment_id; optional content posts a closing reply); action=reopen reactivates it. action=delete PERMANENTLY removes the comment and all its replies (needs comment_id) — deleting is final, resolving is the reversible way to close a thread. Get comment_id from list_comments. Requires a Drive scope on the OAuth token.
- **raw_request** — Escape hatch to call any Google Docs API v1 path directly, for requests the typed tools don't cover — e.g. a batchUpdate with mergeTableCells, updateTableCellStyle, pinTableHeaderRows, updateSectionStyle, updateDocumentStyle, createHeader/createFooter, createFootnote, named ranges, or several requests at once with writeControl.requiredRevisionId: path "v1/documents/<documentId>:batchUpdate", method POST, body {"requests":[...]}. The path may carry a query string. The Bearer token is added automatically; the method defaults to GET. Only docs.googleapis.com paths are reachable — Drive endpoints are not exposed here.

## Claude Desktop Quick Installation
Install path detected from listing signals. Uses `npx` (confidence: high):

```json
"mcpServers": {
  "google-docs-mcp": {
    "command": "npx",
    "args": ["-y","@a1-x-tech/mcp-google-docs@latest"],
    "env": {
      "GOOGLE_DOCS_CLIENT_ID": "",
      "GOOGLE_DOCS_CLIENT_SECRET": "",
      "GOOGLE_DOCS_REFRESH_TOKEN": ""
    }
  }
}
```

**Requires environment variables:** `GOOGLE_DOCS_CLIENT_ID`, `GOOGLE_DOCS_CLIENT_SECRET`, `GOOGLE_DOCS_REFRESH_TOKEN` — the values above are empty placeholders; fill in real credentials before running (see the repository for what each one is for).

## Documentation

## What Google Docs MCP does

Google Docs MCP gives an AI client tools for working with Google Docs through the Google Docs API. It can read documents as text, structure, or Markdown; create documents from Markdown; modify selected ranges; apply text and paragraph formatting; manage lists, tables, breaks, and inline images; work with comment threads; and export documents to formats including Markdown, PDF, and DOCX.

The server is suited to conversational document workflows where an assistant needs to inspect content before making a narrowly scoped change. Typical requests include rewriting a named paragraph, changing heading levels, inserting a table, summarizing unresolved comments, or exporting a document for delivery.

## How it works

Document edits use UTF-16 index positions within a tab body. Insertions and deletions move later positions, so the server directs the assistant to read the document again before each edit and to process edits from the end toward the beginning when appropriate. This approach supports exact range replacement instead of approximate text matching.

Reading a document, its tabs, or its comments does not change Google Docs. Creating a document, inserting or removing content, changing formatting, and managing comments do modify the document. Exporting can write a local file when an output path is supplied while leaving the Google document unchanged.

Markdown conversion works in both directions. A document can be created from Markdown, or its content can be exported as Markdown. Replacing an existing document body with Markdown is an explicitly destructive operation because it replaces the entire body.

## Setup and configuration

Node.js 20 or newer and a Google account are required. Installation does not require credentials at install time. After adding the server to an MCP client, say “connect Google Docs” to start the OAuth flow. The server guides the user through creating an OAuth client, handles Google’s redirect on `127.0.0.1` with PKCE, and retains the tokens itself.

The package can be launched with:

```bash
npx -y @a1-x-tech/mcp-google-docs@latest
```

The documented setup supports Codex, Claude Code, Claude Desktop, Cursor, and VS Code using a local stdio configuration. Codex can also add it with `codex mcp add`, while Claude Code uses `claude mcp add --transport stdio`.

## Tools and capabilities

The server provides 27 tools covering:

- Reading document text, structure, tabs, and Markdown
- Creating documents and inserting or replacing exact ranges
- Styling characters and paragraphs, including headings and lists
- Creating and filling tables
- Adding page breaks and images from public URLs
- Listing, replying to, resolving, and adding document-level comment threads
- Exporting to Markdown, PDF, DOCX, and other supported formats

Images are fetched by Google from a public URL. Supported image types include PNG, JPEG, and GIF.

## Limitations and notes

Tabs can be read and targeted, but the API cannot create, rename, delete, or reorder them. New comments cannot be anchored directly to a text range; they attach at the document level and can optionally quote the relevant text.

Importing Markdown replaces the entire document body. Comment anchors, positioned objects, headers and footers, and additional tabs do not survive that conversion. Export files are limited to 10 MB and exclude comments and suggestions. Inline images require a public URL, support PNG, JPEG, or GIF, and are limited to 50 MB and 25 megapixels. The server does not provide a general-purpose Google Drive tool or an upload channel for image files.

_Full upstream README: https://allmcps.com/mcp/google-docs-mcp/readme_