How does Appwrite support AI agents and MCP?
Appwrite is described as an "MCP & agent-first, OSS platform," and MCP is included by default rather than added on. That means an AI agent can be pointed at a live Appwrite project and act on it directly — for example, querying Postgres, charting traffic, or shipping a campaign — instead of you wiring up a separate integration layer first. This is relevant if you're evaluating whether Appwrite fits an agent-driven workflow; it's less relevant if you only need a traditional backend and don't plan to involve agents.
What "MCP included" actually changes
MCP (Model Context Protocol) is the interface that lets an agent call tools and services on your behalf. Appwrite's positioning is that this interface ships as part of the platform, so the services below are reachable by an agent without you building a custom bridge:
- Auth — email, SMS, OAuth, anonymous sessions, magic URLs
- Databases — Appwrite databases, or managed PostgreSQL and MySQL
- Storage — files with compression, encryption, image transformations, access control
- Functions — serverless functions with isolated runtimes and event-driven execution
- Sites — static, SSR, and CSR frontends deployed from Git with instant previews
- Messaging — email, SMS, and push through one service
- Realtime — subscribe and react to project events as they happen
- Firewall — traffic rules, abuse controls, and edge security per project
The practical effect: an agent operating through MCP sees the same project surface a developer does, so an action like "query the database and send a campaign" maps onto real services rather than a mock.
Which agent tools it works with
Appwrite lists official support for Cursor, Claude Code, and Codex, plus six more. If your team already works in one of those, the integration path is the one the vendor supports rather than a community workaround.
How the AI Arena benchmark helps you choose
Appwrite publishes an "AI Arena" benchmark comparing leading models on how well they operate Appwrite services. The reported figures:
| Model | Cost/1M | Overall | Auth | TablesDB | Functions | Storage | Sites | Messaging |
|---|---|---|---|---|---|---|---|---|
| GPT 5.5 | $5.00 | 97.7% | 98.5% | 96.5% | 99.5% | 94.3% | 100% | 100% |
| Claude Opus 4.7 | $5.00 | 97.1% | 99% | 91.3% | 100% | 94.8% | 100% | 100% |
| Claude Opus 4.8 | $5.00 | 97.1% | 99.3% | 96.1% | 95% | 94.3% | 100% | 100% |
| Grok Build 0.1 | $1.00 | 96.7% | 92% | 96.3% | 100% | 93.5% | 99.7% | 100% |
Read this as a vendor-run comparison, not an independent one. The useful signal is the spread: Storage and TablesDB are where models diverge most, while Sites and Messaging sit at or near 100% across the board. If your agent workflow leans heavily on database operations, that column is the one worth checking before you pick a model.
When this fits your workflow
Appwrite's agent story is a good match if:
- You want agents to act on a live project, not a sandbox — the services above are the same ones you'd use in production.
- You're already using Cursor, Claude Code, or Codex and want a supported integration.
- You want one platform covering auth, data, storage, functions, hosting, and messaging so the agent has a single surface to learn.
It's a weaker fit if you need a backend with no agent involvement, or if you require an independently verified benchmark before committing — the AI Arena numbers come from Appwrite itself.
To see current plan details and any access conditions, check the pricing page; the source material doesn't state what's free or what requires login.