Connecting AI Tools

Scryer’s real power comes from having AI agents work directly with your architecture model. The agent sees the same nodes, edges, and contracts you see in the editor — no lossy text descriptions.

Quick setup

1. Open your model in scryer
2. Click the project menu (top bar) and link your project directory
3. Scryer detects installed AI tools and offers to configure them

That’s it. The next time you start your AI agent in that project directory, it will have access to your architecture model through MCP.

What gets configured

Claude Code

Scryer writes two files:

• .mcp.json in your project root — tells Claude Code where to find the scryer MCP server
• .claude/settings.local.json — auto-approves scryer’s read-only tools so the agent doesn’t prompt you for every get_model call

Codex

Scryer writes:

• .codex/config.toml — MCP server configuration for Codex

Other agents

Any agent that supports MCP can connect. You’ll need to point it at the scryer-mcp binary — check your agent’s MCP configuration docs.

Manual setup

If you prefer to configure things yourself:

Claude Code — add to .mcp.json in your project root:

{
  "mcpServers": {
    "scryer": {
      "type": "stdio",
      "command": "/path/to/scryer-mcp"
    }
  }
}

Codex — add to .codex/config.toml:

[mcp_servers.scryer]
command = "/path/to/scryer-mcp"

You can also run scryer-mcp init from the command line in your project directory — it detects installed tools and writes the config files for you.

Auto-approve read tools

For Claude Code, you can auto-approve scryer’s read-only tools so the agent can browse your model without prompting. Add to .claude/settings.local.json:

{
  "permissions": {
    "allow": [
      "mcp__scryer__list_models",
      "mcp__scryer__get_model",
      "mcp__scryer__get_node",
      "mcp__scryer__get_rules",
      "mcp__scryer__get_changes",
      "mcp__scryer__get_structure"
    ]
  }
}

Scryer can set this up for you automatically from the project menu.

Verifying the connection

Once configured, start your AI agent and ask it to read your model:

“Use scryer to show me the current architecture model”

The agent should call get_model and describe what it sees. If the nodes and edges match what’s on your canvas, you’re good to go.

What agents can do

Through the MCP server, agents can:

• Read the full model, individual nodes, changes since last read, and project structure
• Write nodes, edges, flows, groups, and source mappings
• Get tasks — implementation work ordered by dependencies with inherited contracts
• Validate the model against C4 rules

All changes made by the agent appear on your canvas in real-time.