Explore your connected data
Once your agent is connected to Anything Graph MCP, browse live schema and sample rows before authoring a playbook. Credentials stay in your profile; the agent only sees structure and read-only previews.
Prerequisites
- Anything Graph is running (
anythinggraph start) - Data sources are connected
- Your agent is connected to MCP
- Admin bearer token in MCP config (authoring and introspection tools require admin)
What to ask your agent
Start with a discovery prompt:
Using anythinggraph-cli MCP:
1. list_sources — show my connected data sources
2. For each source, introspect_source to list tables, collections, or objects
3. sample_source on 2–3 interesting resources (limit 5 rows each)
Replace source names with yours (e.g. warehouse_pg, payroll_csv, mongo_main).
MCP tools for exploration
| Tool | Purpose |
|---|---|
list_sources |
Profile keys and adapter types — no secrets returned |
introspect_source |
Tables, columns, foreign keys, or API shapes (read-only) |
sample_source |
A few raw rows from one table, collection, or object |
get_adapter_guide |
How to map this adapter in bindings (call before authoring) |
Example: Postgres
introspect_source with source_id warehouse_pg and schema_name public
Then sample a table:
sample_source with source_id warehouse_pg, resource users, limit 5
Example: CSV
introspect_source with source_id payroll_csv
Column headers come from the file — no schema_name needed.
Example: MongoDB
introspect_source with source_id mongo_main and schema_name mydb
What you are looking for
As you explore, note:
| Question | Why it matters |
|---|---|
| Which tables or files represent people or accounts? | Candidate entities |
| What columns are identifiers vs display fields? | identifier and attributes in the playbook |
| Which foreign keys or shared columns link records? | Relationships in the playbook + link_column in bindings |
| Which sources should sit in one playbook vs separate playbooks? | Federated sources map |
You do not need a playbook to explore. Playbooks come next when you turn this schema into business vocabulary.
Read-only by design
Exploration tools never insert, update, or delete data. sample_source and introspect_source are catalog reads only.
Next step
When you have a mental model of your data, ask your agent to suggest a playbook structure — entities, relationships, and which source each entity uses.