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., "@bundlebuild Build the court bundle from bundle.yaml and flag missing exhibits"
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.
Court bundles from a YAML index. Merge the PDFs, stamp continuous page numbers, generate a hyperlinked index and PDF bookmarks, and pre-flight everything so the missing exhibit is found at your desk rather than at the registry.
title: Bundle of Documents
case: HC/OC 123/2026
parties: Alpha Pte Ltd v Beta Pte Ltd
output: out/bundle.pdf
stamp: { prefix: "AB" } # stamps AB-1, AB-2, ...
sections:
- title: Pleadings
documents:
- { tab: "1", title: Statement of Claim, date: 2026-03-02, file: docs/soc.pdf }
- { tab: "2", title: Defence, date: 2026-03-23, file: docs/defence.pdf }
- title: Correspondence
documents:
- { tab: "3", title: Letter of demand, date: 2025-11-14, file: docs/lod.pdf, pages: "1-2" }$ bundlebuild check bundle.yaml
[WARN ] tab 3: undated; index will show 'Undated'
[WARN ] tab 4: no text layer on any page; OCR before filing
5 documents, 10 pages, 0 error(s), 2 warning(s)
$ bundlebuild build bundle.yaml
wrote out/bundle.pdf: 1 index page(s) + 10 bundle page(s), 5 tabs
index: out/bundle.index.jsonBundle assembly is paid-product territory (Bundledocs, Opus 2, Casedo) and the free
alternative is a paralegal with Acrobat and an afternoon. The mechanics are not hard:
pypdf merges and annotates, reportlab draws. What is missing is a tool that treats the
index as the source of truth and checks the inputs before anyone prints 800 pages.
Related MCP server: benspdf
Step | Detail |
Pre-flight ( | Missing files, non-PDFs, unreadable or encrypted PDFs, page selections out of range, landscape pages, pages with no text layer (scans), undated documents, tabs out of order. Errors block the build; warnings are printed. |
Merge | Documents in manifest order, optional per-document page subsets ( |
Stamp | Continuous page numbers with optional prefix ( |
Index | Generated front matter: court, case, parties, title, volume; then Tab / Document / Date / Page rows grouped by section. Spills across pages as needed. |
Hyperlinks | Every index row is a link annotation to its first page. |
Bookmarks | PDF outline: sections, then |
Sidecars |
|
pip install -e ".[dev]" # library + CLI + tests
pip install -e ".[mcp]" # adds the MCP serverPython 3.11+. Pure Python; no Ghostscript, no Acrobat, no Java.
bundlebuild init [bundle.yaml] # write a template manifest
bundlebuild check bundle.yaml # pre-flight; exit 1 on errors
bundlebuild build bundle.yaml # check, then build; exit 1 on errors
bundlebuild build bundle.yaml --skip-check{ "mcpServers": { "bundlebuild": { "command": "bundlebuild-mcp" } } }Tools: manifest_template, check_bundle(manifest_path), build_bundle(manifest_path).
The assistant's job is to help write the manifest from a folder of documents; the build is
deterministic and local.
Field | Notes |
| Front matter on the index page. Only |
| Relative to the manifest file. |
| First stamped number (default 1; use e.g. 341 for volume 2). |
| Stamp appearance. |
| Default |
| Section heading in the index and the outline. |
| A string label; must be unique across the bundle. |
|
|
| Relative to the manifest. Must be a PDF; convert first. |
| Subset, 1-based, e.g. |
| For authorities bundles; printed in the index and exported to the citecheck sidecar. |
The index is not page-numbered; bundle page 1 is the first content page. If your practice direction requires the index within the numbering, open an issue; it is a small change but it should be a deliberate one.
Existing page numbers on the source documents are not removed. Stamps are placed in the
margin; if a source document already prints in that corner, choose a different
stamp.position.
Rotated source pages are stamped in the page's own coordinate space, so the stamp rotates with the page. This is the correct behaviour for landscape exhibits inserted into a portrait bundle.
No OCR. check tells you which pages lack a text layer; OCR them with whatever you use
(ocrmypdf is good) and re-run.
No conversion from Word/email/images. Convert first, then bundle.
citecheck consumes the authorities sidecar. chronology entries cite bundle page references produced here. See also sg-deadline.
MIT.
This server cannot be deployed
Merge, split, extract, rotate, reorder and stamp PDF pages from your AI chat, all offline.
Merge, split, extract, rotate, reorder and stamp PDF pages from your AI chat, all offline.
Merge, split, extract, rotate, reorder and stamp PDF pages from your AI chat, all offline.
Merge PDFs from your AI chat: combine, order pages, output one clean PDF. All data is local.