Structured tool output
A tool result returned as JSON in structuredContent, besides the text content. A tool that declares an outputSchema must return results that match it.
A plain MCP tool result is a list of content blocks, usually text. That suits a model, but code downstream would have to parse prose. Since the 2025-06-18 revision a tool can also return a JSON object in structuredContent, and can describe its shape in advance with an outputSchema in tools/list. The 2025-11-25 rules are:
- If a tool declares an output schema, servers "MUST provide structured results that conform to this schema", and clients "SHOULD validate structured results against this schema".
- For backwards compatibility, a tool that returns structured content "SHOULD also return the serialized JSON in a TextContent block", so clients that ignore
structuredContentstill see the data.
The 2026-07-28 revision loosens the types: output schemas may use any JSON Schema 2020-12 keywords, and structuredContent may be any JSON value, not just an object.
Example
A weather tool declares "outputSchema": { "type": "object", "required": ["temperature", "conditions"], ... } and answers:
{ "content": [{ "type": "text",
"text": "{\"temperature\": 22.5, \"conditions\": \"Partly cloudy\"}" }],
"structuredContent": { "temperature": 22.5, "conditions": "Partly cloudy" } }A client can now check temperature is a number before a later step relies on it.
Common confusion
The specification is explicit: structuredContent "is server-produced result data and is unrelated to LLM "structured outputs" (schema-constrained model generation)". One describes what a tool returns; the other constrains what a model writes. An output schema is also a contract, and its compatibility rules run the other way from input rules. Making an optional output property required is safe, because consumers only get more; making an optional input argument required breaks every caller that leaves it out.
In L2M
When a tool declares an output schema, L2M checks structuredContent against it. A missing or non-matching object fails the call with a message naming the problem; results flagged isError are not checked, and a schema that cannot be compiled is skipped with a warning. L2M expects an object: an array or a plain value in structuredContent fails the call, even though 2026-07-28 allows it. Later nodes read fields from the MCP Tool node's structured output, and the server registry diffs output schemas with the reversed rules (structured output). Through the MCP gateway, a tool whose output is a JSON object returns it as structuredContent next to the JSON text; the gateway's tools/list declares no output schemas.
Further reading
- MCP specification 2025-11-25: tools, structured content (modelcontextprotocol.io)
- MCP 2025-06-18 changelog (modelcontextprotocol.io)
- MCP 2026-07-28 changelog (modelcontextprotocol.io)