# Get a schema-checked answer out of an agent run

> Get a schema-checked answer out of an agent run: 1 companies fully do it, 15 with a limit, 19 carry the tag. Every listing in Agent frameworks & SDKs with the reason and a source.

- Canonical: https://www.anchorterminal.com/niches/frameworks.typed-output/
- Markdown: https://www.anchorterminal.com/niches/frameworks.typed-output/index.md (~1,800 tokens)
- Slim: https://www.anchorterminal.com/niches/frameworks.typed-output/index.min.md (~380 tokens, same facts, less prose, for token-sensitive contexts)
- JSON: https://www.anchorterminal.com/niches/frameworks.typed-output/index.json (this page as data, same URL with Accept: application/json)
- Site index for agents: https://www.anchorterminal.com/llms.txt (the key docs and every listing in one file, at most 16 MB, about 4 million tokens: https://www.anchorterminal.com/llms-full.txt)
- API: https://www.anchorterminal.com/api/v1/index.json
- MCP: https://www.anchorterminal.com/mcp (the whole catalogue as tools, no key)
- DNS: dig +short TXT <slug>.grade.anchorterminal.com (the grade as a DNS TXT record, for a sandbox with no HTTP where names still resolve; also top., best.<category>. and <a>.vs.<b>. under the zone, and the same under grade.letme.dev)
- Licence: CC BY 4.0, credit Anchor Terminal and link this page (https://www.anchorterminal.com/terms/#data)
- Updated: 2026-11-15

One company's docs show a full fit.

| Companies, full fit | With a limit | Carry the tag | Share of the category's work | First-hand accounts |
| --- | --- | --- | --- | --- |
| 1 | 15 | 19 | 5 | 3 |

## What counts as done

The final output validates against the schema on every run of a fixed test set, invalid output triggers a retry or a typed error rather than a silent wrapper, and tools stay usable in the same run. Docs state which providers support it natively and what happens on the others.

How agents ask for it: return the result as this JSON schema; extract these fields and validate them; give me a typed object, not prose; make the agent's answer match our Pydantic model

## Does the job

The docs show it meets the test above, and nothing we read says otherwise.

- **Pydantic AI (Pydantic Services Inc.)**: output_type validates with Pydantic, by tool output by default, NativeOutput or PromptedOutput, retrying with validation errors, and tools stay usable in the same run.

## Does the job, with a limit

It does the job, but something in the docs or a user's account will get in an agent's way. The limit is named in each row.

- **AgentOS (Framers Lab, Inc.)**: generate() retries up to maxRetries and throws StructuredOutputError, and a provider capability table names native modes. The docs don't show a schema-checked final answer in the same run as the agent's own tools.
- **Agno (Agno Inc.)**: output_schema validates into Pydantic in the same run as tools, with JSON fallback where native schemas are missing. The page says JSON mode doesn't guarantee compliance and names no retry on invalid output.
- **BoxLang AI (Ortus Solutions, Corp.)**: Schemas from classes, struct templates or JSON Schema are validated, with a retry on invalid JSON. Claude structured output needed a tool-forcing workaround (3.3.2) and the schema was once not sent on Bedrock (bx-ai #199, fixed).
- **Claude Agent SDK (Anthropic, PBC)**: outputFormat takes JSON Schema, Zod or Pydantic after multi-step tool use and ends with error_max_structured_output_retries when retries run out. An open account reports output sometimes wrapped in an extra field (claude-agent-sdk-python #502).
- **CrewAI (CrewAI, Inc.)**: Tasks take output_pydantic or output_json alongside tools, and guardrails retry up to guardrail_max_retries. The docs don't say which providers support schemas natively.
- **Docker Agent (Docker, Inc.)**: structured_output takes a schema in native mode (the provider's JSON mode) or tool mode, with strict for native only. The page doesn't state retry on invalid output.
- **Agent Development Kit (ADK) (Google LLC)**: In Python, output_schema and tools work on the same agent and the docs name Gemini on Google Cloud and LiteLLM providers with response schemas as native. Retry on invalid output isn't stated; acc-48's tools-off behaviour is superseded in Python.
- **LangGraph (LangChain, Inc.)**: response_format picks ProviderStrategy where native (OpenAI, xAI, Gemini, Claude) and ToolStrategy otherwise; with tools the model must support both at once. This is LangChain's create_agent on LangGraph rather than LangGraph itself.
- **LlamaIndex Framework (LlamaIndex, Inc.)**: Agents take output_cls (Pydantic) or a structured_output_fn alongside tools. The page doesn't name which providers do this natively or what happens on invalid output.
- **Mastra (Kepler Software, Inc.)**: structuredOutput validates into response.object, optionally through a second model. The page warns some models can't combine tools and structured output and has workarounds.
- **Microsoft Agent Framework (Microsoft Corporation)**: response_format takes a Pydantic model or JSON schema and ChatClientAgent supports it with compatible clients; the page says not all agent types support it natively, and doesn't state retry on invalid output.
- **OpenAI Agents SDK (OpenAI OpCo, LLC)**: output_type with strict schemas and tools in the same run, and an invalid_final_output error handler. The pages read don't state behaviour on non-OpenAI providers.
- **smolagents (Hugging Face, Inc.)**: final_answer_checks can reject a final answer that fails a check, but there is no schema-typed final output; MCP tools can return structured output.
- **Strands Agents (Amazon Web Services, Inc.)**: Structured output returns a validated object; validation errors go back to the agent to correct, a missing object raises StructuredOutputException, and tools work in the same call. The page doesn't say which providers support it natively or how it works on the rest.
- **AI SDK (Vercel Inc.)**: generateText and streamText take output: Output.object({ schema }) alongside tools, counting as an extra step against stopWhen. Retry on invalid output and per-provider native support weren't stated on the pages read.

## Couldn't check

The docs were closed to us, or silent. These carry no mark either way.

- **Cursor Agent SDK (Anysphere, Inc.)**: The pages read show tool input schemas and an outputSchema advertised to MCP clients without validation; no agent-level final output schema was found, and the evals pages weren't opened.

- not this job: guilds.run (Cigale AI): No output schema for a run's final answer in the docs; agents write to notes and collections instead.
- not this job: NitroStack (Nitrostack Inc.): Validates tool inputs with schemas; there is no agent run whose final answer could be schema-checked.
- not this job: TradingAgents (Tauric Research): The manager and trader agents use structured output internally (since 0.2.4), but a caller can't supply its own schema for the final answer.

## What stops agents finishing

- output schema disables tools and hand-offs
- schema silently not sent for one provider, so the model returns prose
- output intermittently wrapped in an extra field

## Around it

Before: frameworks.tool-calling-agent. After: inference.structured-extraction, data.load-rows.

## Where this comes from

Weight 5 of 100 in Agent frameworks & SDKs: 3 of 61 accounts have this as their main job (share 4.9, rounded by largest remainder to sum 100); likely under-counted, since validation retries rarely produce issues

As data: https://www.anchorterminal.com/api/v1/use-cases/frameworks.typed-output.json, or the MCP tool get_use_case with id frameworks.typed-output. Draft, read 2026-11-15 from public docs and first-hand accounts, not yet reviewed by a person.

---

Query this catalogue directly over MCP: https://www.anchorterminal.com/mcp (Streamable HTTP, no key). How: https://www.anchorterminal.com/agents/#mcp
