Tools

MCP tool: validate_resolved

Validate a ResolvedDatachainInstance against its referenced schema.

`validate_resolved`

Soft-failure (isError: false) on every validation outcome. ok: true and ok: false are both successful tool calls — the answer is in the envelope.

Summary

MCP sibling to POST /schemas/:version/validate_resolved. Validates a snapshot-pinned ResolvedDatachainInstance end-to-end:

Element-id resolution against schema_snapshot.elements ∪ suggested_elements (snapshot wins; collision is a hard error).
R14 wire enforcement: non-empty suggested_elements requires authoring_provenance.kind === 'ai_generated'.
Existing semantic rules (cardinality, required categories, context refs) operating against the merged element pool.
Snapshot consistency vs the live store, when the pinned version is still served (R9).

Input

{
  "version": "dtpr@2026-09-01-beta",
  "datachain": { /* ResolvedDatachainInstance */ }
}

The schema_snapshot is bulky and has to agree with the placements, so take the whole datachain object straight from a resolve_datachain response rather than hand-writing it.

Output — valid

{
  "structuredContent": {
    "ok": true,
    "data": { "ok": true, "warnings": [] },
    "meta": { "content_hash": "sha256-…", "version": "dtpr@2026-09-01-beta" }
  }
}

Output — invalid

{
  "structuredContent": {
    "ok": false,
    "errors": [
      {
        "code": "element_id_collision",
        "message": "id 'foo' present in both schema_snapshot.elements and suggested_elements",
        "path": "suggested_elements"
      }
    ]
  }
}

Error codes

Code	Meaning
`parse_error`	Resolved-form Zod parse failed.
`element_id_collision`	Same id in `schema_snapshot.elements` and `suggested_elements` (R15a).
`provenance_required`	Non-empty `suggested_elements` without `authoring_provenance.kind === 'ai_generated'` (R14).
`element_provenance_unknown_element`	A key in `authoring_provenance.element_provenance` matches neither a placement's `element_instance_id` nor a uniquely-placed `element_id`.
`element_provenance_ambiguous`	A key in `authoring_provenance.element_provenance` matches an `element_id` placed more than once. Use `element_instance_id` on each placement and key the entry by that id.
`INSTANCE_ELEMENT_INSTANCE_ID_DUPLICATE`	Two placements share the same `element_instance_id`. Must be unique within `elements[]`.
`INSTANCE_ELEMENT_INSTANCE_ID_COLLIDES_WITH_ELEMENT_ID`	An `element_instance_id` matches an `element_id` placed on this instance. The two namespaces must be disjoint.
`variable_rationale_unknown_variable`	A key inside an `element_provenance[<id>].variable_rationale` map references a variable not declared on the placed element's category.
`unknown_element_id`	Placement element_id absent from snapshot ∪ suggested (R15 fallthrough miss).
`snapshot_drift`	`schema_snapshot` differs from canonical; skipped when the pinned version is no longer in the index.
Semantic validator codes	Shape-valid but semantically wrong (`REQUIRED_CATEGORY_MISSING`, etc.).

Example

# A ResolvedDatachainInstance is too large to hand-write, so resolve a thin
# instance first and feed the result straight back in.
DC='{"id":"acme-resume-screener","schema_version":"dtpr@2026-09-01-beta","created_at":"2026-09-01T00:00:00Z","elements":[
  {"element_id":"employment"},
  {"element_id":"organization","context_type_id":"deployer","variables":[
    {"id":"title","value":[{"locale":"en","value":"Acme Corp"}]},
    {"id":"description","value":[{"locale":"en","value":"Runs the screener."}]}]},
  {"element_id":"people","context_type_id":"subject","variables":[
    {"id":"title","value":[{"locale":"en","value":"Job applicants"}]},
    {"id":"description","value":[{"locale":"en","value":"Anyone who submits a resume."}]}]},
  {"element_id":"recommends","context_type_id":"human_decides"}]}'

RESOLVED=$(curl -s https://api.dtpr.io/mcp -H 'content-type: application/json' \
  --data "$(jq -nc --argjson dc "$DC" '{jsonrpc:"2.0",id:1,method:"tools/call",params:{name:"resolve_datachain",arguments:{version:"dtpr@2026-09-01-beta",datachain:$dc}}}')" \
  | jq '.result.structuredContent.data')

curl -s https://api.dtpr.io/mcp -H 'content-type: application/json' \
  --data "$(jq -nc --argjson r "$RESOLVED" '{jsonrpc:"2.0",id:2,method:"tools/call",params:{name:"validate_resolved",arguments:{version:"dtpr@2026-09-01-beta",datachain:$r}}}')"