2026-04-30 – Claude Goes Creative: Eight Connectors for Adobe, Blender, Ableton & More

🧭 Claude for Creative Work: Eight Connectors Bring AI Into Adobe, Blender, Ableton, and Five More Apps

Anthropic published Claude for Creative Work on April 28, launching eight connectors that give Claude direct access to the creative software millions of designers, developers, and artists use daily. The connectors span 3D modelling, digital audio workstations, visual effects, image editing, vector graphics, and sample libraries — the most comprehensive creative-tooling push any frontier AI lab has made to date. Alongside the product announcement, Anthropic joined the Blender Development Fund as a Corporate Patron at €240,000 per year, putting real money behind the open-source 3D ecosystem it now integrates with.

The eight connectors at a glance

Adobe — 50+ tools across Photoshop, Premiere, and Express. Claude can edit images, cut sequences, and generate assets through natural-language instructions inside Creative Cloud.
Ableton — grounds Claude's answers in official product documentation for Live and Push hardware. Useful for production workflows without leaving the DAW.
Affinity by Canva — automates batch image adjustments, layer renaming, and file exports. Canva's professional-tier suite now has an AI co-pilot.
Autodesk Fusion — creates and modifies 3D models through conversation. Subscribers can describe a mechanical part and have Claude generate a parametric solid.
Blender — exposes Blender's full Python API to Claude. Users can analyse entire scenes, debug object hierarchies, batch-apply modifiers, and add custom tool panels — all in plain English.
Resolume Arena and Wire — real-time control for live visual performance. VJs can trigger clips and adjust composition parameters mid-set via Claude instructions.
SketchUp — converts natural-language descriptions into 3D modelling starting points. Architects and interior designers can rough out layouts without touching the toolbar.
Splice — searches Splice's royalty-free sample catalogue through conversation. Musicians describe the sound they need; Claude surfaces matching stems and loops.

What this means for creative workflows

Each connector is implemented as an MCP server — the same protocol Claude Code uses for tool integrations — so the underlying architecture is consistent whether you are scripting a Blender scene or editing a Photoshop layer. That consistency matters: any wrapper or gateway you build around one connector transfers to the others with minimal modification.

Getting started: Blender Python API via Claude

The Blender connector exposes bpy through Claude's tool-use layer. Once connected, you can ask Claude to "select all objects whose name starts with 'lamp_' and set their emission strength to 5" and it will write and execute the script without you touching a line of Python. This is the fastest way to understand Blender's data model: ask Claude to explain what it did, and you have working sample code with a plain-English annotation attached.

University partnerships included

Anthropic simultaneously announced educational partnerships with the Rhode Island School of Design (RISD), Ringling College of Art and Design, and Goldsmiths University of London. Students and faculty at these institutions get access to Claude alongside the creative connectors — embedding AI literacy into design and art education at the curriculum level rather than leaving it to individual learners.

The Blender Corporate Patron commitment

Joining the Blender Development Fund at the Corporate Patron tier (€240,000/year) puts Anthropic in the same bracket as Epic Games and Meta. The stated focus is the Blender Python API — the layer Claude's connector relies on. From an enterprise risk standpoint this is sensible: by co-funding the API's stability, Anthropic reduces the chance that a Blender upstream change breaks the connector it just shipped to users.

⭐⭐⭐ anthropic.com

🧭 Claude Code v2.1.121 & v2.1.122: alwaysLoad MCP, Bedrock Service Tiers, and /resume PR URL Search

Two Claude Code releases shipped this week — v2.1.121 and v2.1.122 — adding a meaningful MCP configuration option, a new Bedrock infrastructure control, and a quality-of-life improvement to /resume session search. The accompanying bug fix list is the longest in several releases and clears several long-standing irritants, particularly around Vertex AI and structured output.

alwaysLoad: keep critical MCP tools permanently visible

v2.1.121 introduces an alwaysLoad flag on individual MCP server config entries. When set to true, all tools from that server bypass tool-search deferral and are always pre-loaded into the context window, regardless of what the session's ToolSearch mechanism would otherwise decide.

// .claude/settings.json
{
  "mcpServers": {
    "my-internal-db": {
      "command": "npx",
      "args": ["my-db-mcp-server"],
      "alwaysLoad": true
    }
  }
}

When to use alwaysLoad

Use it for MCP servers whose tools your agents call on virtually every turn — for example, a database schema tool or an internal knowledge-base search. Without alwaysLoad, ToolSearch may defer those tools if the session starts with a different focus, causing unnecessary round-trips. Reserve it for servers you know are always relevant; overusing it expands your context budget even when the tools are idle.

ANTHROPIC_BEDROCK_SERVICE_TIER: flex vs priority routing

v2.1.122 adds the ANTHROPIC_BEDROCK_SERVICE_TIER environment variable. Setting it to flex, priority, or default sends an X-Amzn-Bedrock-Service-Tier header on every request. This lets Claude Code sessions participate in the Bedrock capacity tier system — useful for teams that have negotiated priority capacity and want agent sessions to use it automatically rather than competing with the standard queue.

# In your shell profile or CI environment
export ANTHROPIC_BEDROCK_SERVICE_TIER=priority

/resume now accepts PR URLs

Pasting a GitHub, GitHub Enterprise, GitLab, or Bitbucket pull request URL directly into the /resume search box now finds the session that produced that PR. Previously you had to remember part of the session description or branch name. For teams using Claude Code across many parallel feature branches, this removes the friction of rediscovering old context when a PR comes back for review.

Notable bug fixes

Vertex AI structured output — fixed invalid_request_error on structured-output queries through Vertex AI, which had blocked teams using Vertex as their Claude provider for schema-enforced outputs.
Image resize regression — images were being scaled to 2576px instead of the correct 2000px maximum for Opus 4.7 and newer models. Fixed.
MCP tool visibility — ToolSearch was silently missing MCP tools that connected after session start. Those tools are now discoverable mid-session.
Malformed hooks crash — a single malformed entry in settings.json hooks was invalidating the entire file and disabling all hooks. Each hook is now validated independently.
/branch timeline fork — forked branches produced with /branch could fail with a "tool_use ids without tool_result blocks" error when the source session contained rewound timeline entries. Fixed.
OpenTelemetry type correctness — numeric attributes on api_request and api_error log events were being emitted as strings. They are now correctly typed as numbers, which matters for any observability pipeline that groups or aggregates on those fields.

⭐⭐⭐ github.com/anthropics/claude-code