vectrify is a local SVG editor for turning a raster image into editable vector artwork. Alongside manual tools it offers automated operations that work against a reference image: generate new shapes, improve existing ones, and simplify the result. Every operation previews its result and applies it as one undoable edit, limited to the objects and kinds of change you allow.
uv tool install "vectrify[all]" # or: pipx install "vectrify[all]"
vectrify # open the editor in its own window
vectrify drawing.svg --reference original.png
vectrify --serve --port 8765 # or serve it to a browserThe desktop extra (included in all) opens the editor in a native window.
Without it, or with --serve, vectrify serves the editor on loopback and
prints the address to open in a browser. Neither needs a Node build step. From
a source checkout, run uv run vectrify. See the editor guide
for every control.
Vectrify is an MCP server through which an agent (Claude Code, Claude Desktop or any MCP client) looks at a drawing and its reference and edits it with the editor's own commands, one undoable step per call.
From the editor. Turn on Agents in the editor's footer (installed
with [mcp] or [all]): the editor itself hosts the MCP server over
Streamable HTTP at http://127.0.0.1:8770/mcp (the next free port if that
one is taken), and the footer's popover shows the command that adds it to
Claude Code, with a Copy button:
claude mcp add --transport http --scope user vectrify http://127.0.0.1:8770/mcp \
--header "Authorization: Bearer <token>"Add it once: the token is kept across restarts (regenerate it from the same popover), and the client reaches that window whenever Agents is on, so you watch each edit happen.
Headless, over stdio. vectrify-mcp edits a file with no window, or
connect()s to a running editor that allows agents:
claude mcp add vectrify -- vectrify-mcp # installed with [mcp] or [all]
claude mcp add vectrify -- uvx --from "vectrify[mcp]" vectrify-mcpdocs/mcp.md lists the tools and how the live channel works.
| Action | Method | What it does |
|---|---|---|
| Generate | SAMVG | Segments the reference with SAM and traces each region |
| Generate | Colour regions | Fits a colour palette on the GPU and traces its regions |
| Improve | Tidy | Tidies the selected paths in seconds: snaps their points to the reference and simplifies, with Add detail and a gradient shape fit on request; simplifies without one |
| Improve | Fit colours | Closed-form flat fill colours, geometry locked |
| Simplify | Clean up | Drops redundant vertices and merges compatible paths |
Generated shapes are placed over the artboard exactly where the reference is shown. Improve and Simplify change only the selection, and locks, pins and permissions are enforced by the backend for every method. docs/operations.md describes the operation contract for writing new methods.
Python 3.10 or newer. SVG rendering needs Cairo; on Debian/Ubuntu install it
with sudo apt install libcairo2.
The vision and samvg extras install PyTorch and transformers, which SAMVG,
colour regions and the shape fit of Tidy need; all installs both.
Colour regions need an NVIDIA GPU with CUDA; SAMVG uses it when available. The
shape fit runs on the GPU with the optional native CUDA extension (below) and
on the CPU otherwise, except for outlined fills, which need the GPU.
SAMVG is inspired by the SAMVG paper, not an installation of the unreleased research code. It uses SAM ViT-H by default (ViT-B is faster), keeps masks only when they materially improve a flat-colour reconstruction, and traces them into layered SVG paths. It is a general-purpose tracer for photos and painterly images; the editor's Generate dialog chooses the model, the resolution SAM segments at and the maximum number of shapes.
The native CUDA extension for the filled-path fit is built only on request. Build a local wheel with it, then time it:
VECTRIFY_BUILD_SAMVG_CUDA=1 uv build --wheel --no-build-isolation
uv pip install --force-reinstall --no-deps dist/vectrify-*.whl
.venv/bin/python scripts/check_cuda_renderer.pyPyPI releases are portable Python wheels and do not bundle the CUDA extension.
scripts/ holds standalone tools run from a checkout:
bench_colour_regions.py runs colour regions on one image,
bench_samvg_renderer.py and check_cuda_renderer.py time the filled-path
fit, bench_trace.py benchmarks Generate with SAMVG and Tidy on
fixed references, bench_lines.py scores the Cel art tracer's lines
against vector originals (clean, noisy and stretched), bench_shadows.py
measures the shading it leaves out, subtle_screen.py
checks that the scorers in vectrify.score order graded path damage
correctly, and
analyze_profile.py summarises a py-spy profile.