# Directus

Backend and Studio over any SQL database with a built-in MCP server; source-available, free under revenue caps.

Directus wraps a SQL database with REST and GraphQL APIs, a visual Studio for non-technical editors, flows for automation, and policy-based access control. Since v11.12 it includes a native MCP server switched on in Settings > AI, so an agent can manage items, collections, fields, relations, files, and flows under the same permissions as a human user. Configured entirely by env vars. Licensed MSCL 1.0, free for organizations under $5M revenue and 50 staff.

Alternative to: Contentful, Airtable

- Capabilities: adaptable, operable
- Vertical: content-publishing
- Shape: business-app
- Form factor: web-app
- Execution model: hybrid
- Provenance: human-written (recorded, never ranked)
- Repo: https://github.com/directus/directus (MSCL-1.0 (source-available; GitHub NOASSERTION)), 37684 stars, last commit 2026-09-02
- Last activity: 2026-09-02
- Page: apps/directus--directus.html
- JSON: apps/directus--directus.json

## Use it when

- I have a database already and want an admin UI and API on top that my agent can also write to
- I want editors in a Studio and an agent creating collections and content under the same permission rules

## Gotchas

- MSCL 1.0 is source-available, not OSI open source; organizations over $5M revenue or 50 employees need a commercial license
- Free Core tier caps at 3 seats, 25 collections, and 5 flows unless you apply for the Open Innovation Grant
- Remote MCP requires v11.12+; older versions fall back to a Node.js MCP server
- Create a dedicated MCP user and token rather than using an admin token
- 398 open issues; large surface, expect upgrade notes between releases

## Editions

- Core (self-hosted): Free, closed source
  - source: https://github.com/directus/directus
  - container: https://directus.com/docs/self-hosting
  - one-click: https://railway.com/deploy/directus-official
- Open Innovation Grant (self-hosted): Free, closed source
  - hosted: https://directus.com/oig
- Team: $499 per month, closed source
  - hosted: https://directus.com/pricing
- Enterprise: Free, closed source
- Directus Cloud hosting add-on: $99 per month, closed source
  - hosted: https://directus.cloud

## Adaptable: make it yours

Effort: config-only (18 things to edit), surface kind admin-ui

The env surface (.env or CONFIG_PATH: KEY, SECRET, DB_*, PUBLIC_URL, ADMIN_*, storage, LICENSE_*) is deployment configuration; business adaptation lives in the Studio admin UI: Settings > Data Model, Flows, Access Policies and Roles, and Project Settings for branding. Custom email templates need EMAIL_TEMPLATES_PATH plus Liquid files, and anything beyond that is code in extensions/.

- Env vars: KEY, SECRET, DB_CLIENT, DB_CONNECTION_STRING, HOST, PORT, PUBLIC_URL, SERVE_APP, ADMIN_EMAIL, ADMIN_PASSWORD, ADMIN_TOKEN, CONFIG_PATH, LICENSE_KEY, LICENSE_TOKEN, EMAIL_TEMPLATES_PATH
- Config files: .env, docker-compose.yml, extensions/
- Prompt files: none
- Accounts you'll need: Directus (license key for Open Innovation Grant or commercial tiers); OpenAI, Anthropic, or Gemini (optional AI Assistant)

## Operable: what an agent can drive

Operability: official. Interface: mcp at https://directus.com/docs/guides/ai/mcp (published by the builder, both).

### Reach (partial coverage)

- Create, read, update, and delete items: reachable. MCP docs: items create, edit, manage
- Build collections: reachable. MCP docs: build and configure collections
- Define fields and validation: reachable. MCP docs: field types and validation rules
- Model relations: reachable. MCP docs: M2O, O2M, M2M, M2A relationships
- Manage file metadata and folders: reachable. MCP docs: alt text, descriptions, tags, folder organization
- Create automation flows: reachable. MCP docs: generate flows triggered by database events
- Upload binary files: undetermined. metadata management documented; binary upload over MCP not stated
- Manage users, roles, and policies: undetermined. MCP respects permissions; managing them is not listed
- Translate content: undetermined. AI Assistant runs translations in Studio; MCP tool not listed
- Build Studio dashboards and insights: not reachable. Studio UI feature; no MCP tool

## Code stack

- Languages: TypeScript
- Frameworks: Node.js, Vue
- Package manager: pnpm
- Database: PostgreSQL, MySQL, MariaDB, MS SQL, SQLite, OracleDB, CockroachDB
- Hosting target: Docker, Railway one-click, Directus Cloud

## Model stack

- Providers: OpenAI, Anthropic, Google Gemini (AI Assistant, optional)
- Models: none
- Keys you need (names only): none
- Prompts live in: none
- Tools: none

## Evidence and confidence

- formFactor: README: API layer plus visual Studio, self-host or cloud (confidence 0.95)
- executionModel: core is deterministic; README: embedded AI Assistant creates, translates, and takes actions with OpenAI, Anthropic, or Gemini (confidence 0.7)
- operability: README and directus.com/docs/guides/ai/mcp: native MCP server enabled in Settings > AI with a token, published by Directus (confidence 0.9)
- effort: directus.com/docs/configuration/general: env vars in .env through CONFIG_PATH; triage confirmed extensive documented env surface (confidence 0.85)
- editions: directus.com/pricing: Core free with caps, OIG free under thresholds, Team $499 per month annual, Enterprise custom, Cloud $99 per month add-on (confidence 0.8)
- active: pushedAt 2026-09-02
- provenance: Monospace-built since 2004 lineage; no vibe-coded claim (confidence 0.9)
- reach:  (confidence 0.75)

## Open questions

- Full MCP tool names (docs page describes capabilities without listing tool names)
- Does the Core tier's 25-collection cap apply to collections created over MCP?
- Confirm DB_CLIENT and DB_CONNECTION_STRING names on the database configuration page (general page fetched only)

## Sources

- https://github.com/directus/directus (sourcing:content-publishing:vendor)
- https://directus.com/docs/guides/ai/mcp (analysis:content-publishing)
- https://directus.com/docs/configuration/general (analysis:content-publishing)
- https://directus.com/pricing (analysis:content-publishing)
