gdb-regression-test

GDB MCP Regression Test Runner

Execute comprehensive regression tests against a live GDB instance using MCP tools directly.

CRITICAL: All MCP tool calls MUST execute in the main conversation context. Do NOT delegate to subagents — subagents cannot access MCP connections. Every tool invocation must be a direct MCP call from this session.

Workflow

1. Instance Selection

Ask the user which GDB instance to test against. Default: guatemala-dev.

The selected server name becomes the tool prefix: mcp__{server_name}__{tool_name}. Example: mcp__GDB-local-dev__gdb_status.

2. Load Tools

Use ToolSearch to load required MCP tools before each phase:

ToolSearch(query="+GDB-local-dev gdb_status")
ToolSearch(query="+GDB-local-dev gdb_catalog")
ToolSearch(query="+GDB-local-dev gdb_database")
ToolSearch(query="+GDB-local-dev gdb_data")

3. Authentication

GDB uses BPA's authentication. Load and call auth_login from the BPA server:

ToolSearch(query="+BPA-local-dev auth_login")
mcp__BPA-local-dev__auth_login(instance="guatemala-dev", username=..., password=...)

Ask the user for credentials if not known.

4. Execute Phases

Run each phase sequentially. Report results as a table after each phase. Track: phase name, test count, pass/fail, any errors.

If a phase fails, report the failure and ask if the user wants to continue or stop.

Phase 1 — Discovery Mirror (read-only)

Compare MCP read tools against known GDB state on the target instance.

Ground truth for Guatemala dev "Trade names" database:

• catalog_code: trade names, catalog_id: 3, db_id: 4
• Schema fields: ID, Trade name, date 1, Activity, City, Request date, Expiration date, Used, Owner ID, eR User ID, Applicant
• Record count: 7, first record ID: 51

#	Test	Tool	Check
1	Server health	gdb_status(instance)	status == "ok"
2	Instance list	gdb_instance_list()	count >= 1, instance present
3	Catalog count	gdb_catalog_list(instance)	count >= 10
4	Trade names exists	(from catalog_list)	code "trade names" in catalogs
5	Database list	gdb_database_list(instance)	count >= 10
6	Schema fields	gdb_database_get(database_id=4, instance)	11 expected fields present
7	Record count	gdb_data_list(code="trade names", instance)	count == 7
8	Record content	gdb_data_get(data_id=51, instance)	ID == "TRADE NAMES1", City == "Ben Ghazi"
9	Search field	gdb_data_search_field(code="trade names", q="Ben", field="City", instance)	results >= 1
10	Tag list	gdb_tag_list(instance)	has count and tags keys

Phase 2 — Database Lifecycle (write, then cleanup)

Full round-trip: create → verify → publish → CRUD records → cleanup.

#	Test	Action	Check
1	Create database	gdb_database_create(catalog_name="GDB Regression", code="gdbregtest", schema=..., is_draft=True, schema_tags=[{name:"",path:"/Name",is_fulltext:true}], instance)	id > 0, catalog_code == "gdbregtest"
2	Verify exists	gdb_database_get(database_id=<id>, instance)	is_draft == True, Name in schema
3	Schema tags	(from database_get)	schema_tags has /Name entry
4	Publish	gdb_database_publish(database_id=<id>, instance)	returns id
5	Verify published	gdb_database_get(database_id=<id>, instance)	is_draft == False
6	Create record	gdb_data_create(database_id=<id>, content={Name:"Alice",Email:"alice@test.com",Age:30}, instance)	id > 0
7	List records	gdb_data_list(code="gdbregtest", instance)	count == 1, Name == "Alice"
8	Get record	gdb_data_get(data_id=<record_id>, instance)	content matches
9	Update + merge	gdb_data_update(data_id=<record_id>, content={Name:"Alice Updated"}, instance) then gdb_data_get	Name updated, Email preserved (merge!)
10	Search field	gdb_data_search_field(code="gdbregtest", q="Alice", field="Name", instance)	results >= 1
11	Delete record	gdb_data_delete(data_id=<record_id>, instance) then gdb_data_list	count == 0

Schema to use:

{
  "type": "object",
  "properties": {
    "ID": {"type": "string", "primaryKey": true, "$id": 1, "readOnly": true,
      "triggers": [{"conditions": [{"logic": "==", "value": "", "gate": "&&"}],
        "actions": [{"type": "set-value", "value": "{code}{indexNoByCode}", "field_id": 1}]}]},
    "Name": {"type": "string", "$id": 2},
    "Email": {"type": "string", "$id": 3},
    "Age": {"type": "number", "$id": 4}
  },
  "$incrementIndex": 4,
  "required": ["ID"]
}

Save database_id, catalog_code, db_uuid, db_version for subsequent phases.

Phase 3 — Batch Operations

Uses the database from Phase 2 (must be published).

#	Test	Action	Check
1	Batch create	gdb_data_create_batch(code="gdbregtest", version=<db_version>, records=[{Name:"Bob"},{Name:"Carol"},{Name:"Dave"}], instance)	created_count == 3
2	Verify count	gdb_data_list(code="gdbregtest", instance)	count == 3
3	Auto-ID	(from data_list)	All IDs start with "gdbregtest"
4	Upsert new	gdb_data_upsert(code="gdbregtest", version=<db_version>, content={Name:"Eve"}, instance)	has "receive" key
5	Verify count	gdb_data_list(code="gdbregtest", instance)	count == 4

IMPORTANT: Batch tools require numeric version (e.g., "1.0"), NOT "any".

Phase 4 — Error Handling

#	Test	Action	Check
1	Update published DB	gdb_database_update(database_id=<id>, schema={...}, instance)	ToolError raised (406)
2	Nonexistent DB	gdb_database_get(database_id=999999, instance)	ToolError with "not found"
3	Empty create	gdb_database_create(catalog_name="", code="", schema={}, instance)	ToolError raised
4	Batch size limit	gdb_data_create_batch(code="gdbregtest", version=<db_version>, records=[501 items], instance)	ToolError with "exceeds maximum"

Phase 5 — Duplicate & Compare

#	Test	Action	Check
1	Duplicate	gdb_database_duplicate(database_id=<id>, instance)	dup_id != original_id
2	Compare schemas	gdb_database_compare(first_database_id=<id>, second_database_id=<dup_id>, instance)	is_equals == True
3	Dup has no data	(best effort)	May have separate catalog code

Save dup_id for cleanup.

Phase 6 — Cleanup

#	Test	Action	Check
1	Delete duplicate	Get UUID via gdb_database_get, then gdb_database_delete(database_uuid=<uuid>, instance)	deleted
2	Delete main DB	gdb_database_delete(database_uuid=<db_uuid>, instance)	deleted
3	Delete catalogs	gdb_catalog_list(instance), find "gdbregtest*", gdb_catalog_delete each	deleted
4	Verify clean	gdb_catalog_list(instance)	no "gdbregtest" catalogs remain

5. Final Report

Present a summary table:

| Phase | Tests | Passed | Failed |
|-------|-------|--------|--------|
| 1. Discovery Mirror | 10 | ? | ? |
| 2. Database Lifecycle | 11 | ? | ? |
| 3. Batch Operations | 5 | ? | ? |
| 4. Error Handling | 4 | ? | ? |
| 5. Duplicate & Compare | 3 | ? | ? |
| 6. Cleanup | 4 | ? | ? |
| **TOTAL** | **37** | **?** | **?** |

If all pass: "GDB MCP regression passed. Safe to release." If any fail: list failures with details and suggest fixes.