v1.3.0

Fixes from real new-user feedback

Two things tripped up a fresh install on the Node (npx) path — both fixed.

"bun: command not found" / Bun required

• The plugin's copy-paste MCP config snippet used bunx. It now emits npx -y aiconnect-figma-mcp — no Bun needed.
• set_image_fill and batch_ops image embedding used Bun.file(), which throws under plain Node. Switched to Node's fs, so local-image fills work without Bun.

Auto-join failing (having to paste a channel id)

• join_channel with no arguments is now robust: it discovers the plugin's channel via the relay and waits up to ~12 s for the plugin to come online. This fixes the race where the agent tried to connect before the plugin was ready, and the case where a standalone relay was in use.
• You can still pass an explicit code: join_channel("<code>").

Install / upgrade

1. MCP config: npx -y aiconnect-figma-mcp (always pulls latest).
2. Download aiconnect-figma-plugin.zip below, unzip, import manifest.json in Figma desktop → Plugins → Development → Import plugin from manifest.
3. Start a fresh agent session, then just ask it to connect and build — join_channel needs no arguments.

v1.2.0

Zero-friction connect

This release removes the fiddly bits new users hit when connecting.

• No channel code to copy. Call join_channel with no arguments — the server detects the running plugin's channel (it observes it through the built-in relay) and connects automatically. (You can still pass a channel code explicitly for an external relay or multiple plugins.)
• The plugin connects on its own. Once your AI agent is running (which starts the MCP server, which hosts the relay), the plugin panel turns green by itself — no button to press.
• Cleaner standalone relay command. If you ever want to run the relay separately: npx -y aiconnect-figma-mcp relay (replaces the awkward -p form). Normally you never need this — the server hosts the relay in-process.
• Friendlier plugin disconnect message and streamlined README.

Install / upgrade

1. Point your MCP client at the server: npx -y aiconnect-figma-mcp (always pulls latest).
2. Download aiconnect-figma-plugin.zip below, unzip, and import manifest.json via Figma desktop → Plugins → Development → Import plugin from manifest.
3. Run the plugin, then ask your agent to join_channel — no arguments needed.

Builds on v1.1.0 (embedded relay) and the MIT relicense.

v1.1.0

What's new

• Built-in relay — no separate terminal. The MCP server now hosts the WebSocket relay in-process: it binds port 3055 on startup and falls back to an existing relay if the port is already in use. A normal setup no longer requires running the relay yourself.
• Correct standalone relay command. If you do want to run the relay separately, it's npx -y -p aiconnect-figma-mcp aiconnect-figma-relay (it's a bin of the aiconnect-figma-mcp package; plain npx aiconnect-figma-relay 404s). README, the in-plugin hint, and setup output are all fixed.
• Clearer plugin disconnect message — points you at your running MCP server (which provides the relay) instead of the old bun socket command.
• Relicensed to MIT (was AGPL-3.0).

Install

1. Add the MCP server to your client config: npx -y aiconnect-figma-mcp
2. Download aiconnect-figma-plugin.zip below, unzip, and import manifest.json via Figma desktop → Plugins → Development → Import plugin from manifest.
3. Run the plugin, copy the channel id, and call join_channel.

v1.0.0

AIConnect for Figma v1.0.0 — drive a live Figma file from any AI agent, 100% local.

Highlights:

• batch_ops — build a whole page/section in one round-trip
• Design intelligence (keyless, local): apply_brand + brand presets, generate_palette/generate_theme (OKLCH), check_contrast (+auto-fix), suggest_fonts, search_icons/insert_icon (Iconify), search_images (Openverse), fill_realistic_content
• Design tokens: variables (create/bind/set), export_tokens (DTCG/CSS/Tailwind), import_tokens from a local file
• get_css (Dev Mode-equivalent inspect, no paid seat), get_status, get_console_logs, get_page_snapshot
• npx distribution + Node-native relay
• 69 tools. Free for individuals & teams, no restrictions.

Measured: ~1 ms median call latency over localhost, 120/120 calls, 0 failures.