Docs

Browsing the schema

• Toggle the Dictionary drawer from the toolbar's dictionary icon. The canvas stays interactive while it's open.
• Tabs: Tables / Columns / Relationships.
• Each tab supports search and a “Hidden only” filter.

Health checks

Design-time validation that catches errors before they reach the database. Issues surface inline on the canvas, in the entity drawer header, and in the Dictionary's Health tab.

Errors (would fail or silently misbehave)

• Empty table or column name
• Table without a primary key
• Duplicate column names within a table; duplicate table names within the diagram
• FK column type doesn't match the referenced PK
• Identity / AUTO_INCREMENT on a non-integer type
• SQLite AUTOINCREMENT on a non-PK column

Warnings (likely modeling mistakes)

• Same-named columns across tables with no FK between them
• A column named {OtherTable}Id that isn't an FK
• Orphan tables (no relationships)
• FK columns with no covering index
• Identifier matches a dialect reserved word
• Foreign-dialect syntax in a default or expression — SQL Server brackets or GETDATE(), a MySQL backtick — when the diagram targets a different dialect

1-click autofix

Four health checks ship with a Fix button. All four go through the standard mutation pipeline, so ⌘Z reverts.

• fk-no-index — adds a covering single-column index on the FK
• fk-type-mismatch — copies the referenced PK column's type onto the FK
• same-name-no-fk — creates the FK to the matching table's PK
• fk-name-not-linked — creates the FK to the table named in the column ({Table}Id heuristic)

Disable rules per-diagram

Each rule can be silenced design-wide (for example, silence the same-named-FK warning for diagrams full of audit columns). The disable list persists with the diagram.

Pair with Claude (AI design review)

On Pro and Team, Pair with Claude sends your diagram's structure to Claude for a review of the design — naming, normalization, missing keys and indexes, modeling smells. It deliberately ignores SQL syntax and dialect nits; the deterministic health checks above own those.

Findings render as collapsible cards in a right-side drawer and in the Health tab — each one jump-to-able and individually dismissible. Dismissals persist per-diagram and stay suppressed on future runs. Results are cached per diagram and flagged stale when the schema changes. Launch it from the StatusBar.

Create a token

• In the app: avatar menu → Settings → MCP tokens → Create token. Give it a name (e.g. “Claude Code laptop”).
• The raw token is shown exactly once, with copy buttons. Store it in your client's config; ERDone keeps only a hash.
• Tokens are revocable at any time from the same panel — the list shows each token's name, prefix, created date, and last use.

Claude Code

One command, pasting your token in place of <token>:

claude mcp add --transport http erdone https://mcp.erdone.net \
  -H "Authorization: Bearer <token>"

Cursor

Add the server to ~/.cursor/mcp.json (global) or a project .cursor/mcp.json:

{
  "mcpServers": {
    "erdone": {
      "url": "https://mcp.erdone.net",
      "headers": {
        "Authorization": "Bearer <token>"
      }
    }
  }
}

Cursor also interpolates ${env:NAME} inside the header, so you can keep the token out of the file with "Authorization": "Bearer ${env:ERDONE_MCP_TOKEN}".

Codex

Codex (OpenAI CLI) reads remote MCP servers from ~/.codex/config.toml — add the server with a bearer header (TOML, not JSON):

[mcp_servers.erdone]
url = "https://mcp.erdone.net"
http_headers = { "Authorization" = "Bearer <token>" }

To keep the token out of the file, use bearer_token_env_var = "ERDONE_MCP_TOKEN" instead of the http_headers line.

Other MCP clients

Any MCP client that supports remote servers with a custom header works. Add the server to your client's MCP configuration with the generic JSON shape (the in-app MCP tokens panel gives you this ready to paste, with your token filled in):

{
  "mcpServers": {
    "erdone": {
      "url": "https://mcp.erdone.net",
      "headers": {
        "Authorization": "Bearer <token>"
      }
    }
  }
}

Claude Desktop

Claude Desktop's Add custom connector dialog only offers OAuth fields — there's no place to paste a token — so connect through a small local bridge instead. Edit ~/Library/Application Support/Claude/claude_desktop_config.json (on Windows, %APPDATA%\Claude\claude_desktop_config.json) and add:

{
  "mcpServers": {
    "erdone": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-remote",
        "https://mcp.erdone.net",
        "--header",
        "Authorization:${AUTH}"
      ],
      "env": {
        "AUTH": "Bearer <token>"
      }
    }
  }
}

Requires node / npx on your PATH; the bridge is fetched automatically on first launch. Then fully quit Claude Desktop (⌘Q — closing the window isn't enough) and reopen. ERDone's tools appear under the connector menu.

Keep the token in the AUTH environment variable exactly as shown, rather than inlining it into the --header value. The bridge splits arguments on spaces, so an inline "Authorization: Bearer …" arrives malformed — routing it through the variable avoids that.

Technical details

• Endpoint: https://mcp.erdone.net.
• Auth: an Authorization: Bearer erdone_mcp_… header. Tokens are read-only and scoped to the workspaces you belong to.
• Protocol: MCP over Streamable HTTP with stateless POST requests. Protocol versions 2024-11-05 through 2025-06-18 are negotiated automatically — any spec-compliant client works.
• Errors are structured and agent-readable (e.g. a not-found response lists the valid diagram names; an ambiguous name returns the candidates) — agents self-correct instead of dead-ending.
• Latency: every call reads your design live from your GitHub repo. Typical calls are sub-second, with a possible cold start on first use.

Add a new entity for a feature

1. Click + on the toolbar to add a table. Drag it where you want.
2. Click the table to open the entity drawer. Type a name. Add columns (the first column named {TableName}Id auto-promotes to PK).
3. To link it to an existing table: drag from the right edge handle of the parent table's PK column onto the body of your new table. ERDone creates the FK column (typed to match) AND a covering index automatically.
4. Open Migration Preview → Save to repo → run via Flyway/etc.

Add a column to an existing table

1. Click the table → entity drawer opens → click the column-add row at the bottom.
2. Name it, set the type. If you named it {OtherTable}Id and there's no FK yet, Health will flag it — click the Fix button to wire the FK automatically.
3. Migration will emit ALTER TABLE … ADD COLUMN.

Change a column's type or nullability

1. Click the table → find the column → change type / toggle NOT NULL.
2. Status bar shows “1 pending.”
3. Migration Preview Changes tab shows: Customer.Email: type: VARCHAR(50) → VARCHAR(120).
4. If you change your mind on a single attribute among many edits, click the ↩ next to that one row to revert just that change.

Refactor a relationship

1. Click the relationship line on the canvas → relationship drawer opens.
2. Re-aim source or reference column via the dropdowns (no need to delete + re-draw).
3. Toggle identifying / cardinality / parent-required / on-delete-cascade as needed.
4. Migration emits DROP CONSTRAINT FK_… then ADD CONSTRAINT FK_… FOREIGN KEY ….

Find and fix unindexed FK columns

1. Open Dictionary → Health tab.
2. Filter for “FK column has no covering index” warnings.
3. Click Fix on each — one click adds IX_{Table}_{Column} covering the FK.
4. Open Migration Preview to see the resulting CREATE INDEX statements.