GuidesProject types
agent
The agent preset: metadata-driven objects with REST and MCP access for AI agents.
create-weavekit-app my-app --type=agentNarrative: AI-Agent backend — metadata-driven objects with REST access for AI agents.
What it scaffolds
- No script subsystem — hooks are off.
features.fieldTypeslimited to the primitives. Opt-in types (relations, currency, enums, custom types) are gated off until you widen the whitelist.- REST (
/api) and MCP (/mcp) enabled, same as every type.
What you'll usually add
- MCP identities for the users agents act on behalf of (
adapters.mcp.identities). RBAC decides each identity's tool surface, so set this up before exposing anything. - Guardrails once agents can write: rate limits, and
requireApprovalfor risky calls. - More field types in
features.fieldTypeswhen the domain needs them.
Read next
- MCP — expose objects as tools, per identity
- Custom tools, guardrails & audit replay
- Approvals — human-in-the-loop for tool calls
- Practices: integrate an existing CRM, design an agent-friendly schema