Suggest a playbook structure

Turn raw schema into a business vocabulary: entity names, relationships, and routing across your connected sources. Your agent can draft this structure from introspection results — you review and refine before saving anything.

Prerequisites

• Explore your connected data — know your tables, collections, or files
• Admin MCP token connected

The goal

A playbook structure answers:

1. Entities — What are the business objects? (crm_user, crm_account, crm_payroll_record)
2. Relationships — How do they connect? (owns_account, user_has_payroll)
3. Sources — Where does each entity live? (postgres, csv, salesforce, …)

Bindings come later. At this stage, focus on names and meaning, not column mappings.

Prompt your agent

I've connected warehouse_pg (Postgres) and payroll_csv (CSV).

Using anythinggraph-cli MCP:
1. introspect_source for both sources
2. Propose a playbook structure for sales users who own CRM accounts and have payroll history
3. Output compact playbook JSON with:
   - entities (identifier + attributes)
   - relationships (from / to)
   - sources map (which entity uses postgres vs csv)
4. Do NOT save yet — show me the draft for review

Adapt the domain in the prompt to your data — healthcare patients, support tickets, inventory SKUs, or whatever you connected.

Use suggest_bindings

After you have a draft playbook id and entities, the agent can call suggest_bindings to heuristically map entities to tables or files:

For playbook_id my-crm-access:
- get_playbook_context
- suggest_bindings for source warehouse_pg

Review suggestions against what you saw during exploration. The agent should correct mappings where column names differ from playbook field names.

Federated vs single-source

Pattern	When to use
Single source	All entities in one database — one source key (e.g. postgres)
Federated	Entities span systems — multiple source keys, multiple binding files

Example federated sources map:

"sources": {
  "crm_user": "postgres",
  "crm_account": "postgres",
  "crm_payroll_record": "csv"
}

See Example 2: CRM + payroll for a full federated walkthrough.

Review checklist

Before moving on, confirm:

• [ ] Entity names are stable business terms, not raw table names
• [ ] Each entity has a clear identifier field
• [ ] Relationships are directional (from subject → to object)
• [ ] Every entity has a sources entry
• [ ] Scope is intentional — one playbook per access domain, not one giant graph

Next step

Decide who may see what — define access rules (ReBAC) before saving the playbook.