Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
82 changes: 82 additions & 0 deletions apps/web/content/blog/canva-mcp-server-design-automation-guide.mdx
Original file line number Diff line number Diff line change
@@ -0,0 +1,82 @@
---
title: "Canva MCP Server: Create Designs From Claude in 2026"
description: "Canva's official hosted MCP server lets Claude create, edit, and export designs through natural language. Covers setup, OAuth, and which features need a paid plan."
excerpt: "Canva's official MCP server runs at mcp.canva.com/mcp with no local install required. This guide covers what it can create and edit, how OAuth setup works, and which features like Resize and Brand Templates are gated behind paid or Enterprise plans."
date: "2026-07-31"
updatedAt: "2026-07-31"
author: "Gus Marquez"
authorUrl: "https://www.linkedin.com/in/gustavoamarquez"
authorSameAs:
- "https://www.linkedin.com/in/gustavoamarquez"
tags:
- "mcp"
- "beginner"
- "canva"
- "design"
- "media"
category: "media"
focusKeyword: "canva mcp server"
draft: false
cornerstone: false
topic_source: new_topic
cluster: media
faqItems:
- question: "Do I need to know how to code to use the Canva MCP server?"
answer: "No. It's a hosted, remote server, so there's nothing to install or configure beyond authorizing Canva through OAuth in your AI client. If you can describe the design you want in a sentence, you can use it."
- question: "Is the Canva MCP server the same as Canva's developer MCP servers for building apps?"
answer: "No, and it's an easy mix-up. Canva ships three separate MCP-related docs pages: the design-creation server at mcp.canva.com/mcp covered here, a Connect API dev server for building integrations, and an Apps SDK dev server for building Canva apps. Only the first one creates or edits designs directly."
- question: "Can Claude publish a Canva design directly to social media through MCP?"
answer: "The server handles export in formats like PDF, PNG, JPG, PPTX, and MP4, which you can then post yourself. It doesn't include direct publishing integrations with social platforms as of this writing."
- question: "Does every team member need their own Canva account to use the MCP server?"
answer: "Yes. Authentication runs per user through an individual OAuth flow, so everyone connecting an AI assistant to Canva needs their own Canva login, not a shared team credential."
---

Describe a design in one sentence. Canva's MCP server can build that design and export it, handling the edits in between, all without you ever opening the Canva editor. Canva builds and hosts this server itself, and it targets a reader most of MCPFind's developer guides don't: marketing and social teams who just want a design assistant, without any API to wire into their stack. It sits in MCPFind's [media category](/categories/media), which right now indexes 517 servers, most of them design and image tooling. New to MCP entirely? The [what-is-mcp guide](/blog/what-is-mcp) is the better first stop. Below I walk through what the server actually does and where the paid-plan limits kick in.

## What Can the Canva MCP Server Actually Do?

Canva documents the hosted server at canva.dev/docs/mcp, and you reach it at mcp.canva.com/mcp. It manages your brand kit and assets while creating and editing designs straight from a prompt. Library search is baked in as well. Export runs across multiple formats, and it can leave collaborative comments too. All of it runs through natural language instead of the editor's drag-and-drop interface.

In practice, you can ask for a social post in your brand colors, have Claude pull assets from your existing Canva library rather than starting on a blank canvas, and get the result back as a PDF, PNG, JPG, PPTX, or MP4, whichever you need. The commenting tools go a step further: an AI assistant can join a design review thread and leave feedback the way a teammate would, so you're not relaying notes back and forth by hand. If your team splits work between visual marketing assets and developer-facing design-to-code handoffs, it's worth setting this next to [Figma's MCP server](/blog/figma-mcp-server-design-to-code), since the two really do serve different audiences despite both being design tools.

Media's [top-starred entry](/servers/ai-smithery-kirbah-mcp-youtube) is a YouTube server, not a design tool at all. That gap tells you how broad "media" runs on MCPFind: video platform integrations, image generation, and the design tooling this post covers, all filed under one umbrella.

## How Do You Set Up the Canva MCP Server?

You don't install anything locally or generate an API key, and there's no config file to hand-edit, which makes this one of the simpler setups covered on MCPFind. Because it's a hosted remote server, setup comes down to pointing your AI client at the mcp.canva.com/mcp endpoint and completing an OAuth authorization in your browser the first time you use it.

Every person on your team needs their own Canva account and their own OAuth authorization. There's no shared team credential that covers everyone in one shot. Plan around that if you're rolling this out to a marketing team instead of setting it up solo, because each seat authorizes on its own and no admin toggle covers the whole team at once. Once you're connected, ask Claude to list your recent Canva designs. It's a quick way to confirm the authorization actually went through before you try building anything new.

## Which Canva MCP Features Require a Paid Plan?

Connecting to the server is free on any plan, Canva Free included. A few individual capabilities aren't. The Resize tool needs a paid Canva plan, and both Autofill and Brand Templates require Canva Enterprise specifically.

A solo user on the free tier can still create, edit, and export designs through Claude without hitting a paywall for the basics, which is why that split matters more than a blanket "free" or "paid" label suggests. What gets reserved for Enterprise is the higher-leverage automation, the kind that scales one brand template across dozens of variations. If you're still weighing whether this is worth adopting, test the free-tier features first, before you assume an upgrade is the only way to get value out of it.

The free tier suits one-off design work, while the paid tiers are built for repeatable production. A single social graphic or a presentation slide works fine with no upgrade at all. Generating fifty on-brand variants of one template for a campaign launch is where Autofill and Brand Templates start saving real time, and that's the piece Canva gates behind Enterprise specifically. A mid-tier plan won't unlock it.

## How Is Canva's MCP Server Different From Its Developer-Focused MCP Tools?

Canva publishes three separate MCP-related docs pages that sound alike but do very different things, and only one of them creates or edits designs.

The design server at canva.dev/docs/mcp is the one this guide covers, and it's what you want when you're asking Claude to build or change an actual Canva design. A separate Connect API dev server, installed locally with `npx @canva/cli mcp`, exists purely to feed AI coding assistants documentation context while you build integrations against Canva's Connect API. The third, the Apps SDK dev server, does the same documentation-context job but for developers building Canva Apps. Neither of those last two touches your designs, so if you land on Canva's developer docs hunting for "the MCP server" and wind up installing a local npm package instead of authorizing an OAuth connection, you've probably grabbed one of the other two by mistake.

## Frequently Asked Questions

### Do I need to know how to code to use the Canva MCP server?

No. It's a hosted, remote server, so there's nothing to install or configure beyond authorizing Canva through OAuth in your AI client. If you can describe the design you want in a sentence, you can use it.

### Is the Canva MCP server the same as Canva's developer MCP servers for building apps?

No, and it's an easy mix-up. Canva ships three separate MCP-related docs pages: the design-creation server at mcp.canva.com/mcp covered here, a Connect API dev server for building integrations, and an Apps SDK dev server for building Canva apps. Only the first one creates or edits designs directly.

### Can Claude publish a Canva design directly to social media through MCP?

The server handles export in formats like PDF, PNG, JPG, PPTX, and MP4, which you can then post yourself. It doesn't include direct publishing integrations with social platforms as of this writing.

### Does every team member need their own Canva account to use the MCP server?

Yes. Authentication runs per user through an individual OAuth flow, so everyone connecting an AI assistant to Canva needs their own Canva login, not a shared team credential.

This is the first MCPFind post to cover Canva specifically, and one of the first written for non-developer teams rather than the engineers who usually wire up a local server. If your team works across both design and code, pair it with the [Figma MCP server guide](/blog/figma-mcp-server-design-to-code) for the developer-facing half of the same workflow. And browse MCPFind's broader [media category](/categories/media), where you'll find adjacent tools like screenshot-capture servers.
Original file line number Diff line number Diff line change
@@ -0,0 +1,87 @@
---
title: "Home Assistant MCP Server: 4 Options Compared for Claude"
description: "No single official Home Assistant MCP server exists yet. Compare voska/hass-mcp, homeassistant-ai/ha-mcp, achetronic/hass-mcp, and ganhammar's custom component."
excerpt: "Home Assistant has no vendor-official MCP server, so four community projects fill the gap, each with a different architecture. This guide compares voska/hass-mcp, homeassistant-ai/ha-mcp, achetronic/hass-mcp, and a custom-component option so you can pick the one that matches your setup."
date: "2026-07-29"
updatedAt: "2026-07-29"
author: "Adam Bush"
authorUrl: "https://www.linkedin.com/in/adam-bush/"
authorSameAs:
- "https://www.linkedin.com/in/adam-bush/"
tags:
- "mcp"
- "beginner"
- "home-assistant"
- "smart-home"
- "self-hosted"
category: "other"
focusKeyword: "home assistant mcp server"
draft: false
cornerstone: false
topic_source: new_topic
cluster: other
faqItems:
- question: "Is there an official Home Assistant MCP server from Nabu Casa?"
answer: "No standalone Nabu Casa server exists for this. Home Assistant core ships built-in Model Context Protocol integrations that let Home Assistant act as an MCP target for other tools, but no single Nabu Casa-branded MCP server exists for connecting Claude or another AI agent to your smart home."
- question: "Which Home Assistant MCP server should beginners start with?"
answer: "voska/hass-mcp is usually the easiest starting point. It's Python, MIT licensed, runs as an external process, and has the most documentation of the community options, including Docker and uvx install paths."
- question: "Can an AI agent control smart home devices, not just check their status?"
answer: "Yes. All four servers covered here expose service calls, not only entity state reads. So Claude can flip on a light, adjust a thermostat, or fire off an automation, rather than only reporting the current state of one."
- question: "Do any of these Home Assistant MCP servers work remotely, not just on a local network?"
answer: "achetronic/hass-mcp supports HTTP transport with OAuth and JWT validation specifically for remote use, and homeassistant-ai/ha-mcp can be reached through Nabu Casa's cloud or a reverse proxy since it runs inside Home Assistant itself. voska/hass-mcp is built primarily for local and Docker-based setups."
---

Ask Claude to turn off the living room lights or check whether the garage door is closed, and none of it works until a Home Assistant MCP server is sitting between the two. Unlike most vendor integrations MCPFind covers, there's no single official server here. Home Assistant itself ships an MCP integration for exposing your instance to other tools, but wiring an AI agent up to actually control it? That's a job four separate community projects have taken on, each with its own architecture and its own tradeoffs. MCPFind's [other category](/categories/other) is where projects like this usually land when they don't fit a narrower bucket, and it's worth a browse if you're hunting for adjacent home-automation tooling. New to MCP as a concept? The [what-is-mcp guide](/blog/what-is-mcp) covers the basics before you pick a server. This guide compares the four real options so you're not guessing which one fits your setup.

## Why Isn't There One Official Home Assistant MCP Server?

Home Assistant is open source and famously modular. No single company owns the "official AI integration" the way PagerDuty or Stripe owns theirs. Home Assistant core does ship its own Model Context Protocol integrations, but those expose Home Assistant as a target for other MCP-aware tools. They aren't a ready-to-run server that plugs Claude into your smart home out of the box.

The community moved in to fill that gap quickly. Four projects now chase the same basic goal, and they get there by very different routes. voska/hass-mcp runs as an external Python process, the easiest of the bunch to spin up. Then it splits. One project lives inside Home Assistant itself as an add-on, another is a custom component you install through HACS, and the fourth breaks from Python entirely, written in Go and built with production deployment in mind rather than a quick local hookup. None of the four covered here are indexed in MCPFind's directory with meaningful star counts yet, since enrichment for niche self-hosted tools like these tends to lag well behind their real GitHub activity. A related [community entry](/servers/io-github-jarahkon-hass-mcp-server) does already exist in MCPFind's own index, though it's early enough that its star count hasn't caught up to its actual usage either, which is the usual story for the smart-home tools scattered across the directory.

## Which Home Assistant MCP Server Should You Actually Use?

Here's how the four options break down by architecture, language, and how you'd reach them.

| Server | Architecture | Language | Best For |
|---|---|---|---|
| [voska/hass-mcp](https://github.com/voska/hass-mcp) | External process | Python | Local setups, easiest to start with |
| [homeassistant-ai/ha-mcp](https://github.com/homeassistant-ai/ha-mcp) | In-process HA add-on | Python | Nabu Casa cloud or reverse-proxy remote access |
| [achetronic/hass-mcp](https://github.com/achetronic/hass-mcp) | External process | Go | Production deployments needing OAuth and Kubernetes |
| [ganhammar/hass-mcp-server](https://github.com/ganhammar/hass-mcp-server) | HA custom component | Python | Teams already comfortable with HACS custom components |

voska/hass-mcp sits at roughly 300 GitHub stars and ships under an MIT license, which makes it the most established of the four by community adoption. It exposes entity state, service calls, history and statistics, automation listings, and dashboard generation, and it installs through Docker or uvx. homeassistant-ai/ha-mcp goes the other way. It runs as an in-process add-on inside Home Assistant itself rather than a separate service, which requires Home Assistant 2026.6.0 or newer but simplifies remote access through Nabu Casa's existing cloud connection.

## How Do the Go and Custom-Component Options Differ?

achetronic/hass-mcp is the outlier here. It's written in Go instead of Python, and built explicitly for production use. It supports both stdio for local connections and HTTP with OAuth and JWT validation for remote ones, ships a Dockerfile and a Helm chart, and aims itself squarely at teams running Home Assistant behind real infrastructure rather than on a spare Raspberry Pi.

ganhammar/hass-mcp-server isn't a standalone server at all. It's a Home Assistant custom component, installed the same way you'd add any HACS integration, exposing MCP over HTTP transport from inside your existing Home Assistant install. If your team already manages Home Assistant through custom components and HACS, that consistency might matter more to you than whether the underlying code is Python or Go. If it doesn't, one of the standalone servers is probably the simpler pick, since you're not tying the integration's lifecycle to Home Assistant's own update cycle.

## What Can Claude Actually Do Once It's Connected to Home Assistant?

All four servers expose service calls rather than read-only access, so the baseline capability is the same across the group. Claude can read entity state, call services to control devices, pull history and statistics, list existing automations, and in voska's case, generate dashboards from a description.

So both "is the front door locked" and "turn off every light downstairs" run over that same connection, once you've scoped the entities and services you want exposed. It pays to start narrow here. Home Assistant setups often carry dozens or hundreds of entities, and handing an AI agent access to all of them at once, particularly anything tied to locks or garage doors, is a bigger blast radius than most people want on day one. Scope the integration to a handful of rooms or device types first, confirm the responses match what you expect, then widen it once you trust the setup.

Long-term automations are where this really starts to pay off beyond one-off commands. Rather than writing a Home Assistant automation yourself for "dim the lights after sunset if someone's home," you can describe the behavior to Claude, have it read your existing entities and automations for context, and draft the automation logic for you to review before it goes live, and that review is where the real safety net lives. Treat anything Claude proposes as a draft, not a finished automation, especially early on while you're still confirming it understands your entity names and room layout correctly.

## Frequently Asked Questions

### Is there an official Home Assistant MCP server from Nabu Casa?

No standalone Nabu Casa server exists for this. Home Assistant core ships built-in Model Context Protocol integrations that let Home Assistant act as an MCP target for other tools, but no single Nabu Casa-branded MCP server exists for connecting Claude or another AI agent to your smart home.

### Which Home Assistant MCP server should beginners start with?

voska/hass-mcp is usually the easiest starting point. It's Python, MIT licensed, runs as an external process, and has the most documentation of the community options, including Docker and uvx install paths.

### Can an AI agent control smart home devices, not just check their status?

Yes. All four servers covered here expose service calls, not only entity state reads. So Claude can flip on a light, adjust a thermostat, or fire off an automation, rather than only reporting the current state of one.

### Do any of these Home Assistant MCP servers work remotely, not just on a local network?

achetronic/hass-mcp supports HTTP transport with OAuth and JWT validation specifically for remote use, and homeassistant-ai/ha-mcp can be reached through Nabu Casa's cloud or a reverse proxy since it runs inside Home Assistant itself. voska/hass-mcp is built primarily for local and Docker-based setups.

This is the first MCPFind post in this particular grouping of self-hosted, non-vendor smart-home tooling. If you're browsing more broadly for what else MCP can connect to, the [MCPFind blog index](/blog) covers everything from IDE setup guides to vendor deep-dives across every category in the directory.
Loading
Loading