Skip to main content

pac-vector

Local stdio MCP server: an SVG editor over in-memory virtual documents (one per "window", like Inkscape), with a live browser preview of every edit.

  • Content goes in as SVG markup; structure (layers, groups, masks, z-order) has typed tools.
  • Layers are real Inkscape layers, so exported files open in Inkscape with layers intact.
  • Path booleans (union, difference, intersection, exclusion, division) via Skia PathOps.
  • Every edit is atomic (a failed call changes nothing), one undo/redo step, and autosaved: windows of a crashed session reopen in the next one.
  • render returns a PNG the agent can look at; rendering is resvg (no browser, no Cairo).

Install

Same shape as pac-pincers: a pip package, run as a module by whatever Python it was installed into.

python -m pip install --upgrade pac-vector
claude mcp add pac-vector --scope user -- python -m pac_vector

As a plugin .mcp.json entry (mirrors the datatopia plugin's pac-pincers entry):

"pac-vector": {
  "type": "stdio",
  "command": "${PAC_VECTOR_PYTHON:-python}",
  "args": ["-m", "pac_vector"],
  "env": {}
}

Development:

python -m venv .venv
.\.venv\Scripts\python.exe -m pip install -e ".[dev]"
.\.venv\Scripts\python.exe run_tests.py -q
.\.venv\Scripts\python.exe -m build          # dist/pac_vector-<ver>.tar.gz + .whl

Tools (21)

Group Tools
Documents doc_create (blank, file, markup, or duplicate an open window), doc_list, doc_set (size, units, viewBox, background, fit), doc_export (svg/png, page/drawing/ids, plain), doc_close
Reading inspect (outline or xml), render (PNG, optional crop to ids + bbox)
Editing add (markup → layer/group, defs auto-routed; SMIL animations into any element), update (attrs/style/text/transform/label/hidden/locked/text-on-path on many ids), replace (one element or the whole drawing), remove, arrange (z-order, move between layers, duplicate, offset), import_svg (file / other window / markup → editable group or embedded image, fitted into a box)
Structure group (also new layers), ungroup, mask (mask or clip from id or markup), unmask
Geometry boolean (union, difference, intersection, exclusion, division), align (align edges/centers, distribute evenly or by gap; relative to selection, page, first, last or an element)
History undo, redo

Preview

doc_create / doc_list / inspect print preview: http://127.0.0.1:<port>/#<doc>. The page has a tab per document, outlines the elements each edit touched, and shows (and copies) the id of anything you click. The server lives inside the MCP process: it starts on the first doc_create and dies with the session.

Environment

Variable Default Meaning
PAC_VECTOR_PORT 0 (random) Pin the preview port (falls back to random if busy)
PAC_VECTOR_OPEN_BROWSER off 1 opens the preview in the default browser on first start
PAC_VECTOR_ROOT server cwd Base directory for relative export/import paths
PAC_VECTOR_LOG WARNING Log level (stderr only)
PAC_VECTOR_AUTOSAVE on 0 disables the crash-recovery mirror
PAC_VECTOR_AUTOSAVE_DIR <tmp>/pac-vector-autosave Where the mirror lives (<pid>/<doc>.svg + .json)

Notes

  • stdout is the JSON-RPC pipe: nothing may print to it. tests/test_stdout_safety.py scans the package for print() / sys.stdout, and tests/test_stdio.py runs a real session.
  • Native modules are loaded before the stdio loop starts (render.warm): on Windows a first native load from inside the FastMCP loop can hang (see pac-pincers' _excel.py).
  • Bounding boxes are measured from pixels (two-pass render), so they include strokes and filters and work for text and curves; content further than two page-widths off-canvas isn't found. align uses them, so it aligns what you see.
  • Booleans use fill geometry only (strokes, text and images don't take part); the result is a <path> in the bottom operand's place, styled like it, keeping its id.
  • Animations are SMIL: the browser preview plays them; render / PNG export show the static first frame.
  • Version lives only in pyproject.toml; a release is a commit named after the version.

Metadata

Release files for pac-vector 0.2.0

For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.

Source distribution (sdist)

Source distribution for pac-vector 0.2.0
File Size Uploaded Provenance Actions
pac_vector-0.2.0.tar.gz 50.3 kB Without provenance Details for pac_vector-0.2.0.tar.gz

Built distribution (wheel)

Table of built distributions (wheels) for pac-vector 0.2.0
File Interpreter ABI Platform Actions
pac_vector-0.2.0-py3-none-any.whl Python 3 none any Details for pac_vector-0.2.0-py3-none-any.whl

Total release size: 97.4 kB

Release files / pac_vector-0.2.0.tar.gz

Download URL pac_vector-0.2.0.tar.gz
Size 50.3 kB
Tags Source
SHA-256 checksum
How to use checksums
c777cc39f3ef9d3c4894f845aca23e8d1962791fe3ca27b055880743389ff100
BLAKE2b-256 checksum
How to use checksums
4cd874947f3b478764ce6a82c9a7e0510d484e3b8a94b688044ec8ed3f6bbef3
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
No
Uploaded via twine/7.0.0 CPython/3.14.0

Release files / pac_vector-0.2.0-py3-none-any.whl

Download URL pac_vector-0.2.0-py3-none-any.whl
Size 47.1 kB
Tags Python 3
SHA-256 checksum
How to use checksums
2ea3d10a3cd1a982c9bfc086fa00fc2a77cceb1d8b560f037bca0a2d0a6709a3
BLAKE2b-256 checksum
How to use checksums
4501745907dd8e8394d043431bb9cc8cdd3ceb8daff2eb558f070b9e8f156ecc
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
No
Uploaded via twine/7.0.0 CPython/3.14.0

Security for pac-vector 0.2.0

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

No provenance found

None of the files in this release were uploaded with provenance metadata.

A note on package security

PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.

To learn more about protecting yourself from malicious actors, visit our package security documentation.

Anthropic, PBC Visionary sponsor Bloomberg Visionary sponsor Hudson River Trading Visionary sponsor Meta Visionary sponsor NVIDIA Visionary sponsor Microsoft Sustainability sponsor Depot Continuous Integration AWS Cloud computing and Security Sponsor Datadog Monitoring Fastly CDN Google Download Analytics Sentry Error logging StatusPage Status page