r7-office
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@r7-office Convert report.docx to PDF and check it's valid"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
dsh-r7-office
R7-Office (Р7-Офис) document processing plugin and Model Context Protocol (MCP) server for DeepSeek Harness.
Lets an AI agent inspect, read, create, format, edit and convert DOCX, XLSX,
PPTX and PDF documents — through standard OOXML manipulation plus the R7
converter you already have installed — without emulating a mouse or keyboard.
⚠️ Unofficial community project
This is an independent, community-maintained project. It is not affiliated with, endorsed by, sponsored by, or supported by АО «Р7» (R7-Office) or DeepSeek, and it is not an official R7-Office or DeepSeek product.
R7-Office and Р7-Офис are trademarks of their respective owners and are used here only to describe what this software interoperates with.
This repository contains no R7-Office code, binaries or other assets. It detects an R7-Office installation already present on the user's machine and drives it locally. You must install and license R7-Office yourself.
Русская документация: README.ru.md
Related MCP server: docx-mcp-server
What it does
Format-preserving editing — replacements and edits keep the original XML styles, fonts, colours and numbering hierarchies. Untouched parts of a document are preserved byte for byte.
All four formats — DOCX (headings, paragraphs, lists, tables, page breaks), XLSX (sheets, ranges, values, formulas), PPTX (slides, titles, text frames), PDF (conversion through the local R7
x2tengine).Validation built in —
r7_validatechecks package integrity before and after a change.Two ways to run — as a native DeepSeek Harness plugin, or as a standalone MCP server over stdio for any MCP client.
Live desktop bridge — optionally drives a document the user has open in R7-Office Desktop, over a loopback WebSocket.
Architecture
DeepSeek Harness ──► dsh-r7-office plugin ──┐
├──► DOCX / XLSX / PPTX / PDF
Any MCP client ────► MCP server (stdio) ────┤
│
R7-Office Desktop ◄── desktop bridge ◄──────┘Details in docs/architecture.md; design decisions in docs/decisions/.
Requirements
Node.js | >= 20.0.0 to run the plugin and MCP server. >= 22.0.0 to run the full test suite (the bridge and CDP clients use the global |
Runtime dependencies | none — the plugin has zero runtime dependencies |
R7-Office Desktop | optional. Needed for |
Platform | Windows, Linux and macOS. R7 auto-detection covers the standard install locations of all three; the live desktop bridge is verified on Windows only (see limitations). |
Clean install from scratch
These steps assume an empty directory and a machine with Node.js 20+.
# 1. Get the code
git clone https://github.com/blazar-source/dsh-r7-office.git
cd dsh-r7-office
# 2. Install the optional dev dependencies (test-only: the MCP client SDK)
npm install
# 3. Verify the checkout — no R7-Office required for this step
npm testnpm test is the portable suite and needs nothing but Node: unit tests,
OOXML round-trip regression tests, file end-to-end workflows, security-policy
tests and an external MCP client smoke suite. Tests whose subject is the
installed product — R7's own templates, the x2t converter, the desktop editor
— skip themselves with a stated reason instead of failing, so a machine
that has never seen R7-Office gets a green run and a visible skip count.
$ npm test
# tests 469
# pass 390
# fail 0
# skipped 79The R7-dependent tests are only meaningful where R7-Office is installed, so they get their own command, which refuses to run without it rather than skipping everything and reporting success:
# Requires an installed R7-Office. Runs the whole suite with nothing skipped,
# then drives the live desktop editor over CDP and saves a document for real.
npm run test:r7
# Just the suite, or just the live editor, if you want them separately:
npm run test:r7 -- --suite
npm run test:r7 -- --liveTo see what the portable suite does on a machine without R7 even though you
have one installed, set R7_OFFICE_DISABLED=1 — detection then reports no
installation, which is exactly the condition CI runs under.
Then verify the integration you actually intend to use:
# R7-Office file pipeline (author → edit → validate → PDF). Skips if R7 is absent.
npm run test:live
# DeepSeek Harness plugin activation. Boots a fresh Harness and asserts the
# 28 r7_* tools reached the tool registry.
npm run test:harness -- --profile webAs a DeepSeek Harness plugin
Install the package directory as a bundle:
# in the DeepSeek Harness UI: plugin_manager → install_bundle
# target: /absolute/path/to/dsh-r7-officeor from the CLI equivalent for your profile:
dsh plugin --profile <profile> add /absolute/path/to/dsh-r7-officeConfirm it activated — a fresh Harness boot prints:
[r7-office] зарегистрировано инструментов: 27 (r7_inspect, r7_read, ...); desktop bridge port=7888, developerMode=falseNote. Add the row either through
install_bundleor by hand in the profile'scordis.patch.yml— never both. Two entries with the samer7-officeid make the row fail to activate.
The plugin also registers a short usage section into the agent system prompt,
so the agent knows the tools and the intended inspect → read → edit → validate → convert order.
As a standalone MCP server
The MCP server speaks line-delimited JSON-RPC 2.0 on stdio.
{
"mcpServers": {
"r7-office": {
"command": "node",
"args": ["/absolute/path/to/dsh-r7-office/src/mcp/cli.js"]
}
}
}Put that in your client's configuration file (for example
claude_desktop_config.json or .cursor/mcp.json). Any stdio MCP client
works; interoperability is verified in npm test against the official
@modelcontextprotocol/client SDK.
You can also run it by hand:
echo '{"jsonrpc":"2.0","id":1,"method":"tools/list"}' | node src/mcp/cli.jsTools
Tool | Purpose |
| Document outline: headings, paragraphs, tables, sheets, slides, metadata |
| Structured text, Markdown view, or spreadsheet cell ranges |
| New DOCX / XLSX / PPTX. Refuses to replace an existing file unless |
| Replace, restyle or delete one paragraph by index |
| Find and replace text, preserving run formatting |
| Insert paragraphs, headings, bullet items or page breaks |
| Create, inspect or update tables and cells, including merge, borders, shading and column widths |
| Read the normalized formatting of every paragraph, run and table cell: style, font, size, colour, alignment, indents, spacing, line spacing, lists |
| Read and set page size, orientation, margins, columns, page breaks and section breaks |
| List, read, create or retitle headers and footers; page-number fields survive |
| Insert a PNG/JPEG/GIF at a size, keeping the aspect ratio |
| List, insert, retitle or remove hyperlinks and their relationships |
| Read values, formulas or the normalized formatting of a sheet or range (e.g. |
| Write cells or a 2-D matrix, addressed by sheet name or index; dates are stored as real date serials |
| Format a cell or range: font, background, borders, alignment, wrap, number format (integer, decimal, currency, percent, date, datetime, custom), merge, column width and row height |
| Add a worksheet to an existing workbook; other sheets are untouched |
| Insert or update a formula |
| Read a slide as a normalized structure: every object's id, type, geometry, text, font, fill, stroke, alignment and paragraphs |
| Create a deck, or append a slide built on one of the deck's own layouts, without altering the slides already there |
| Restyle or reposition one existing object in place: font, geometry, fill, border, alignment, lists, spacing, text |
| Duplicate, move, reorder or delete slides |
| Add or remove an object: shape, text box or PNG/JPEG image |
| Convert via the R7 |
| Check package integrity and XML health |
| Report whether a PDF really contains text or was produced without fonts: per-font embedded flag and ToUnicode/Cyrillic map counts, text glyphs, empty fill operators, and a |
| Desktop bridge connection and effective security mode |
| Read or replace the selection in the open editor |
| Run a safe editor command, or raw DocScript in developer mode |
Presentations
A deck is built on the layouts it already has. Adding a slide registers the
slide part, its relationship part, its [Content_Types].xml override, its
<p:sldId> entry and the presentation relationship; the layout, the master,
the theme, the notes and every slide that already existed keep their original
bytes. A placeholder written onto a slide stays a placeholder, so it keeps
inheriting its geometry and typography from the master.
Reading resolves that inheritance for you: r7_slide_read reports a title
slide's 60 pt layout title rather than the presentation's 18 pt default, and
reports the position a placeholder inherits from the layout or the master
instead of null.
import { PptxEngine } from 'dsh-r7-office/r7'
const pptx = new PptxEngine()
// 1. Read: geometry, text, font, fill, stroke, alignment, paragraphs.
const before = await pptx.readSlide('Deck.pptx', 1)
const title = before.slide.objects.find(o => o.placeholder?.type === 'title')
const body = before.slide.objects.find(o => o.placeholder?.type === 'body')
console.log(title.x, title.width, title.font.family, title.font.size)
// 2. Add a slide on one of the deck's own layouts.
const added = await pptx.addSlide('Deck.pptx', {
layoutType: 'obj',
title: 'Ключевые выводы',
paragraphs: [{ text: 'Выручка +18%', bullet: true }]
})
// 3. Restyle one object. Everything not named keeps its original bytes.
await pptx.formatObject('Deck.pptx', {
slideIndex: added.slideIndex,
objectId: title.id,
font: { family: 'Georgia', size: 32, bold: true, color: '#1F6FEB' },
alignment: 'center'
})
// 4. Add an object: fill, border, text, font, geometry — all in one call.
await pptx.addShape('Deck.pptx', {
slideIndex: added.slideIndex,
shape: 'rounded-rectangle', // rectangle, ellipse, line, arrow, star, …
x: '2cm', y: '10cm', width: '9.4cm', height: '8.2cm',
fill: '#1F6FEB', fillTransparency: 0.1,
line: '#0B3D91', lineWidth: 2, // lineWidth is in points
text: 'KPI 98%',
font: { family: 'Arial', size: 24, bold: true, color: '#FFFFFF' },
alignment: 'center', verticalAnchor: 'middle'
})
// 5. Insert and replace a picture.
const image = await pptx.addImage('Deck.pptx', {
slideIndex: added.slideIndex, imagePath: 'chart.png', x: '4cm', y: '4cm', width: '24cm'
})
await pptx.formatObject('Deck.pptx', {
slideIndex: added.slideIndex, objectId: image.objectId, imagePath: 'chart-v2.png'
})
// 6. Structure, and a PDF.
await pptx.duplicateSlide('Deck.pptx', 1)
await pptx.moveSlide('Deck.pptx', 2, 0)
await pptx.deleteSlide('Deck.pptx', 4)
console.log((await pptx.validate('Deck.pptx')).valid)
await pptx.toPdf('Deck.pptx', 'Deck.pdf')Run the full acceptance deck:
node examples/pptx-acceptance.js # writes into os.tmpdir()/r7-acceptance
node examples/pptx-acceptance.js ./out # or wherever you likeMeasurements. A bare number is EMU (the unit r7_slide_read returns), so a
value can be read, adjusted and written back unchanged. "2cm", "1in",
"30px" and "24pt" also work. lineWidth is the exception: a number below
100 is read as points, because "border: 1.5" means 1.5 pt to everyone who is
not holding a DrawingML specification.
Colours. #RRGGBB, #AARRGGBB and transparency as a separate 0..1 option.
transparency on a font and fillTransparency on a fill are deliberately
distinct, so a half-transparent caption colour cannot make the shape under it
see-through.
Shapes. rectangle, rounded-rectangle, ellipse, circle, line,
arrow (and arrow-left/up/down/left-right), triangle, diamond,
pentagon, hexagon, octagon, star, chevron, plus, cloud, heart,
cylinder, cube, donut, pie, parallelogram, trapezoid — or any
DrawingML preset name. pptx.shapeCatalog() lists them all.
Typical agent flow
"Take
Report.docx, update section 3, keep the formatting, add a summary table and save a PDF."
r7_inspect → r7_read → r7_replace (keeps formatting) → r7_table
→ r7_validate → r7_convertRun it yourself:
node examples/report-scenario.jsLibrary use:
import { DocxEngine, R7Adapter } from 'dsh-r7-office/r7'
const docx = new DocxEngine()
const adapter = new R7Adapter()
await docx.replaceText('Report.docx', 'draft text', 'approved text', {
outputPath: 'Report_v2.docx' // the original is never overwritten by default
})
await docx.table('Report_v2.docx', {
action: 'create',
rows: [['Objective', 'Timeline', 'Owner'], ['Deploy R7', 'Q2', 'IT']]
})
console.log((await docx.validate('Report_v2.docx')).valid)
await adapter.convert('Report_v2.docx', 'Report_v2.pdf')Security
r7_desktop_exec can execute code inside the user's open editor. Raw
DocScript is therefore disabled by default; only a fixed allowlist of safe
argument-driven commands runs in production. Enable raw execution only
deliberately:
- id: r7-office
name: 'dsh-r7-office'
config:
developerMode: true # or export DSH_R7_DEVELOPER_MODE=1r7_desktop_status always reports the effective mode. The desktop bridge binds
to 127.0.0.1 only.
See SECURITY.md for the full threat model and how to report a vulnerability.
Known limitations
Presentations: building layouts or masters is out of scope. Slides are created on the layouts the deck already contains; a deck that ships none gets the generic title/body pair.
Presentations: SmartArt and charts are preserved, never authored. The reader reports them (
type: "chart",type: "graphicFrame") and every part behind them stays byte-identical through any edit, but the engine cannot create one. A hand-written SmartArt frame is just adgm:relIdsreference, and R7's own renderer dereferences the diagram parts behind it.Presentations: image replacement stores a fresh media part when the format changes or the media is shared, so the old part can be left unreferenced. Everything that still points at it keeps working.
Presentations: theme colours read back as tokens (
scheme:accent1), because no literal hex exists until the theme is resolved. Writing accepts literals only.Live desktop bridge: Windows verified. The bridge plugin itself is platform-neutral, but the automated live test drives R7-Office Desktop through the CEF DevTools protocol and is only verified against the Windows build (
Editors-2026.3.1). Other platforms should work; they are untested.R7 required for PDF.
r7_convertneeds a local R7 installation for itsx2tconverter. Everything else works without it.Format coverage. Reading and editing cover the common OOXML surface listed above. Charts, embedded objects, SmartArt and tracked changes are preserved but not editable through these tools.
DOCX: not implemented in v0.1.0. Replacing or resizing an existing image; assigning a table style (
tblStyle); changing a hyperlink's target (retitle, or remove and insert instead); a convenience wrapper for per-section different headers (both can be read and preserved, andr7_docx_header_footeracceptsfirst/even— forfirstalso callr7_docx_sectionswithtitlePg). Search is not revision-aware:r7_replacescans raw<w:t>, so text inside a tracked insertion is editable and paragraph indices count paragraphs inside deleted content —<w:delText>is never matched.DOCX: comments, tracked changes, footnotes, endnotes, a table of contents, equations and embedded objects are preserved, never edited. A test asserts they survive an ordinary edit with every part byte-identical.
DOCX to PDF: a thin horizontal line can appear through an italic subtitle. A paragraph styled
Subtitle(italic, centred, grey) may render with a faint line across it afterr7_convert. The document contains no strikethrough, underline or paragraph border —w:strike,w:uandw:pBdrare all absent, and the same style applied to R7's own template renders cleanly through the same converter — so the line is introduced during PDF conversion and cannot be corrected from the document. It is cosmetic: the text itself is correct and extracts normally. Quantified at roughly 5 % more ink in that text band.PDF rendering is verified visually, not by text extraction. A PDF can carry the right text, the right page count and a clean
validate()while drawing the wrong glyphs or a solid block.scripts/visual-acceptance.pyrenders the output with two independent engines and compares against a LibreOffice render of the source; run it when changing anything in the conversion path.XLSX: not implemented in v0.1.0. Cell styles are limited to the properties listed for
r7_sheet_format; conditional formatting, data validation, charts, pivot tables and defined names are preserved but not editable.PPTX: SmartArt, charts, animations and transitions are preserved, never authored. The engine cannot create SmartArt (a hand-written frame is only a
dgm:relIdsreference), and editing their content is out of scope.No concurrent editing. File tools operate on a document at rest. For a document the user has open, use the desktop bridge; the file tools assume the file is not locked by another process.
Development
npm test # the portable suite; needs only Node, R7 tests skip
npm run test:unit
npm run test:integration
npm run test:e2e
npm run test:r7 # requires R7-Office: full suite + the live desktop editor
npm run test:live # just the live desktop bridge (needs R7-Office)
npm run test:harness # boots a fresh DeepSeek Harness
npm run inspect:r7 # CDP probe of a running R7 DesktopSee docs/development.md and docs/roadmap.md.
License
MIT — see LICENSE.
R7-Office is proprietary software owned by АО «Р7». This project redistributes none of it.
Available Tools
28 toolsr7_convertA
Convert document to PDF, HTML, TXT, DOCX, XLSX, or PPTX using R7 native converter (x2t). XLSX to PDF exports EVERY worksheet by default, in tab order; pass sheetName (or allSheets: false) to export one sheet only.
Inputs: sourcePath (required), targetPath (required), allSheets, sheetName.
r7_createA
Create a NEW DOCX, XLSX or PPTX document. Refuses to replace an existing file unless overwrite is true.
Inputs: filePath (required), title, sheets, tables, +2 more.
r7_desktop_execB
Execute safe built-in command or custom DocScript in live opened document in R7 Desktop. Arbitrary JS requires developerMode enabled.
Inputs: args, code, safeCommand.
r7_desktop_selectionC
Get or replace selected text in the currently active R7-Office Desktop editor window.
Inputs: action (required), text.
r7_desktop_statusA
Check live connection status and security policy for open R7-Office Desktop windows via the local bridge.
Inputs: no named parameters.
r7_docx_formattingA
Read the normalized formatting of a DOCX document before changing it: per-paragraph style, font family/size/bold/italic/underline/strike/colour, alignment, indents, spacing, list numbering, plus table cell styles (widths, merges, shading, borders, alignment), page/section setup and header/footer parts. Values are normalized (points, #RRGGBB), never raw OOXML.
Inputs: filePath (required), count, includeRuns, fromParagraph, +4 more.
r7_docx_hyperlinkA
Read, insert, re-text or remove DOCX hyperlinks. Reading returns each link's text and target URL; inserting writes the relationship (external, deduplicated by URL) and a Word-compatible Hyperlink-styled run; re-texting keeps the relationship and the run formatting.
Inputs: filePath (required), url, bold, size, +12 more.
r7_docx_imageA
Insert a PNG/JPEG/GIF into a DOCX document, or list the images already in it. Inserting creates the media part, its relationship and its content type, and keeps the picture's own aspect ratio when only one dimension is given. Existing images and relationships are never touched.
Inputs: filePath (required), alt, data, name, +8 more.
r7_docx_sectionsA
Read or change DOCX page setup and sections: page size, portrait/landscape, margins, columns, page-number restart, page breaks and section breaks. Changing one property preserves everything else in the section (headers, footers, columns, document grid, title-page setting).
Inputs: filePath (required), page, type, action, +13 more.
r7_editA
Edit, update text, style, or remove a specific paragraph in a DOCX document. The paragraph keeps its own properties (section break, list numbering, spacing) and the new text inherits the previous first run's formatting unless you ask for different formatting.
Inputs: filePath (required), paragraphIndex (required), style, newText, +4 more.
r7_insertC
Insert a paragraph, heading, list item, or page break at the start, the end, or relative to another paragraph. The inserted paragraph can carry its own alignment, indents, spacing and font.
Inputs: filePath (required), bold, list, runs, +22 more.
r7_inspectC
Inspect document structure: outline, metadata, headings, tables, sheet list, slide list, sections, headers/footers, images and hyperlinks.
Inputs: filePath (required), includeLists, includeImages, includeStyles, +3 more.
r7_pdf_inspectA
Measure the text fidelity of a PDF: page and font inventory, how many glyphs are really drawn by text operators, how many glyph fills painted nothing, whether the text is extractable, and a verdict (text/outlined/mixed). Use it to prove a PDF export kept its text instead of silently losing it; set repair to fix malformed ToUnicode CMap entry counts that make R7-exported body text copy out as garbage.
Inputs: filePath (required), repair.
r7_readB
Read structured text, filtered paragraphs, markdown preview, spreadsheet ranges or slides, optionally with normalized formatting.
Inputs: filePath (required), count, query, range, +7 more.
r7_replaceB
Search and replace text in DOCX while strictly preserving all existing run formatting, fonts, colors, and styles.
Inputs: search (required), replace (required), filePath (required), matchCase, +2 more.
r7_sheet_addA
Add a new worksheet to an existing XLSX workbook, optionally with initial data. Other sheets are left untouched.
Inputs: filePath (required), data, name, index, +1 more.
r7_sheet_formatA
Apply formatting to a cell or range in an XLSX worksheet: font, fill, borders, alignment, number format, merge, column width and row height. Values, formulas and unrelated styles are preserved. A sheet's print layout (fit-to-width so a table does not spill onto another page, orientation, paper, margins, centring) can be written in the same call with pageSetup.
Inputs: filePath (required), fill, font, merge, +11 more.
r7_sheet_formulaC
Insert or update a formula in an XLSX spreadsheet cell.
Inputs: cell (required), formula (required), filePath (required), sheetName, +3 more.
r7_sheet_readB
Read spreadsheet values, formulas and formatting from a worksheet or range (e.g. A1:D10).
Inputs: filePath (required), range, sheetName, sheetIndex, +2 more.
r7_sheet_writeC
Write values, dates or formulas into an XLSX worksheet, addressed by index or name.
Inputs: filePath (required), cells, matrix, sheetName, +3 more.
r7_slide_createA
Create a new PPTX deck, or append a slide to an existing one built on one of the deck's own layouts. Appending never modifies the slides already present, and the layout, master and theme the deck already has are reused rather than replaced. Pass baseSlideIndex to clone an existing slide instead of using a layout.
Inputs: filePath (required), title, subtitle, layoutName, +5 more.
r7_slide_editA
Change the structure of a PPTX deck: duplicate a slide, move a slide, reorder all slides, or delete a slide. Slides are addressed by their position in the deck, and no slide part is rewritten by a reorder, so nothing on any slide can be affected.
Inputs: action (required), filePath (required), order, toIndex, +4 more.
r7_slide_formatA
Format or reposition one object on a PPTX slide, in place. Sets font (family, size, bold, italic, underline, colour), geometry (x, y, width, height, rotation, z-order), fill and border, paragraph alignment, bullet and numbered lists, line and paragraph spacing, and can replace the object's text. Only the properties you name change: gradients, shadows, hyperlinks and everything else the author set are left untouched. Falls back to the slide's first text object when no target is named.
Inputs: filePath (required), x, y, bold, +43 more.
r7_slide_objectA
Add or remove objects on a PPTX slide: shapes (rectangle, rounded-rectangle, ellipse/circle, line, arrow, triangle, star, chevron, plus more), text boxes and PNG/JPEG images. An object gets its geometry, fill, border, text and font in the same call. Existing media and relationships are never damaged. Set action "replaceImage" with an existing image object's id to swap its picture while keeping its geometry and placement.
Inputs: action (required), filePath (required), x, y, +47 more.
r7_slide_readA
Read a PPTX slide as a normalized structure: every object with its id, type, current geometry (x, y, width, height, rotation, z-order), text, font, fill, stroke, alignment and paragraphs. Placeholder geometry and typography are resolved from the slide layout and master, so inherited values are reported rather than left null. Also returns the deck's layouts, so the next slide can be built on a real one.
Inputs: filePath (required), includeRaw, slideIndex, includeInherited.
r7_tableB
Create, inspect, or modify tables in DOCX documents: add and remove rows and columns, merge and unmerge cells, set column widths, and format individual cells (shading, borders, alignment, width). Cell formatting never rewrites the text or the properties of neighbouring cells.
Inputs: action (required), filePath (required), cell, rows, +16 more.
r7_validateB
Validate document package structure, XML integrity, and formatting health.
Inputs: filePath (required).
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 28 tools
The set includes both generic readers (r7_read, r7_inspect) and format-specific readers (r7_sheet_read, r7_slide_read, r7_docx_formatting) that read overlapping content, creating ambiguity about which tool to use for a given read task. Descriptions help, but several tools target similar resources.
Names follow a prefix r7_ and mostly snake_case, but mix verb-first (r7_sheet_add) with noun-first (r7_docx_formatting) and generic verbs (r7_read) alongside domain-scoped ones, so the pattern is inconsistent though readable.
28 tools is on the high side for a single server, exceeding the typical 3-15 range; while the broad domain (DOCX, XLSX, PPTX, PDF, desktop) could justify many operations, the count feels heavy and could overwhelm.
Core document CRUD/formatting is well covered across DOCX, XLSX, PPTX, and PDF, including tables, sections, headers, images, hyperlinks, slides, and conversion. Minor gaps exist (e.g., worksheet delete/rename, DOCX style management, PDF editing), but agents can work around most.
Maintenance
Related MCP Connectors
Convert, create and edit Word, PowerPoint, Excel and PDF documents from any AI agent.
Online office suite for AI agents: edit docs, sheets, slides and PDFs in browser or via MCP.
Document conversion and OCR for AI agents: PDF, Office docs, images to text.
Turns agent output into versioned web pages, PDFs, Word docs, spreadsheets and Markdown.
Related MCP Servers
- AlicenseAqualityDmaintenanceEnables AI agents to read, write, and edit Office documents via LibreOffice with token-efficient design. Supports multiple formats including DOCX, XLSX, PPTX, and legacy formats through LibreOffice bridge.2723 npmMIT
- AlicenseNot gradedqualityDmaintenanceEnables AI agents to read, edit, and create Microsoft Word documents (.docx) with support for rich text, tables, and images, deployable locally or via SSE.3MIT
- AlicenseBqualityCmaintenanceEnables complete Office document lifecycle management for AI agents, including creation, editing, conversion, and templating of DOCX, XLSX, PPTX, PDF, and EML files.4055 PyPIMIT
- AlicenseAqualityDmaintenanceEnables AI agents to generate real Office documents (.pptx, .docx, .xlsx) and source code from natural language via MCP protocol.4MIT