Anthropic Claude

Use Anthropic Claude models with Agno agents.

Install the Anthropic model extra, set an API key, and pass Claude to an Agent.

Installation

uv pip install -U "agno[anthropic]"

Agno v3.0 requires anthropic>=0.77.0, and the extra pins it. That release is the first with the output_config request parameter, which Agno now uses for structured output; Agno also routes temperature, top_p, and top_k through extra_body, which is how the Claude models keep working on anthropic 1.0.0, where output_format and those sampling parameters were removed from the SDK’s message methods. If you pin the SDK yourself, raise the floor: on an older SDK, a request that sets output_config fails at send time with TypeError: ... unexpected keyword argument 'output_config'; plain requests are unaffected.

Authentication

Set the ANTHROPIC_API_KEY environment variable. Create a key in the Claude Console.

Mac/Linux

export ANTHROPIC_API_KEY="your_anthropic_api_key"

Example

agent.py

from agno.agent import Agent
from agno.models.anthropic import Claude

agent = Agent(
    model=Claude(id="claude-sonnet-4-5-20250929"),
    markdown=True,
)

if __name__ == "__main__":
    agent.print_response("Write a two-sentence horror story.")

Save the example as agent.py, then run:

python agent.py

See the basic usage guide for the complete environment setup.

Model Selection

Claude defaults to claude-sonnet-4-5-20250929 in Agno v2.7.2. Pass another model ID with Claude(id=...) when a different capability or latency profile fits the use case.

Model	Use Case
claude-sonnet-4-5-20250929	Balanced speed and intelligence
claude-opus-4-5	Complex agentic coding and enterprise work
claude-haiku-4-5-20251001	Fast, cost-sensitive workloads
claude-sonnet-4-6	Long-running agent workflows

See Anthropic’s current model comparison and model lifecycle table before deploying a pinned model.

Claude 4.5 model IDs use dated snapshots. Shorter names such as claude-sonnet-4-5 are aliases. Starting with Claude 4.6, dateless IDs such as claude-sonnet-4-6 are canonical pinned versions, not rolling aliases.

Supported Input

Input	Agno v2.7.2 Support
Text	Text input and text output
Images	JPEG, PNG, GIF, and WebP from a URL, local path, or bytes. Anthropic analyzes only the first frame of an animated GIF.
PDFs	PDF documents from a URL, local path, bytes, or Anthropic file ID
Text documents	UTF-8 content, including .txt, .csv, and .md, from a URL, local path, bytes, or Anthropic file ID
Audio and video	Unsupported. The adapter omits these inputs and logs a warning.

Agno sends image URLs to Anthropic as base64 data after downloading them. Anthropic’s request-size, image-size, page-count, and model limits still apply. See the vision limits and PDF limits.

Request Limits

The Messages API requires max_tokens. Agno sends 8192 unless max_tokens is set on Claude. Anthropic also accepts 0 for prompt-cache pre-warming, but Agno v2.7.2 omits that falsy value; use a positive value with this adapter. See the Messages API reference.

Anthropic also applies organization and workspace rate limits. See the rate-limit documentation.

Beta Features

Pass exact Anthropic beta names with betas. Features that add request fields also need their matching Agno parameters. This context-editing configuration sends both the required beta and request body:

from agno.agent import Agent
from agno.models.anthropic import Claude

agent = Agent(
    model=Claude(
        id="claude-sonnet-4-5-20250929",
        betas=["context-management-2025-06-27"],
        context_management={
            "edits": [{"type": "clear_tool_uses_20250919"}],
        },
    ),
)

See beta features with Claude and context editing.

Prompt Caching

Set cache_system_prompt=True to add a prompt-cache breakpoint to the agent’s system prompt:

from agno.agent import Agent
from agno.models.anthropic import Claude

agent = Agent(
    model=Claude(
        id="claude-sonnet-4-5-20250929",
        cache_system_prompt=True,
    ),
)

See prompt caching with Claude.

Structured Outputs

Pass a Pydantic model as output_schema to use Claude’s native structured outputs. Anthropic supports structured outputs on Claude Opus 4.1, the Claude 4.5 families, and later models. Claude 3.x, Claude Sonnet 4, and Claude Opus 4 do not support the feature.

Agno v3.0 sends structured output through Anthropic’s output_config.format request shape, which requires anthropic>=0.77.0 (the agno[anthropic] extra pins it). Earlier Agno versions used the legacy output_format request, which anthropic 1.0.0 removed.

from agno.agent import Agent
from agno.models.anthropic import Claude
from pydantic import BaseModel


class User(BaseModel):
    name: str
    age: int
    email: str


agent = Agent(
    model=Claude(id="claude-sonnet-4-5-20250929"),
    description="Extract user information.",
    output_schema=User,
)

See Anthropic’s structured-output compatibility and the Agno guides:

• Basic structured outputs
• Streaming structured outputs
• Structured outputs with strict tools

Params

Parameter	Type	Default	Description
id	str	"claude-sonnet-4-5-20250929"	Anthropic Claude model ID
name	str	"Claude"	Model name
provider	str	"Anthropic"	Model provider
max_tokens	Optional[int]	8192	Maximum output tokens. Agno v2.7.2 sends only truthy values, so Anthropic’s 0-token cache pre-warming is unavailable through this adapter.
thinking	Optional[Dict[str, Any]]	None	Model-specific thinking configuration
output_config	Optional[Dict[str, Any]]	None	Anthropic output configuration forwarded unchanged, such as {"effort": "medium"}
temperature	Optional[float]	None	Sampling temperature when supported by the selected model
stop_sequences	Optional[List[str]]	None	Strings that stop generation
top_p	Optional[float]	None	Nucleus-sampling value when supported by the selected model
top_k	Optional[int]	None	Top-k sampling value when supported by the selected model
cache_system_prompt	Optional[bool]	False	Add cache control to the agent-built system prompt
extended_cache_time	Optional[bool]	False	Use a one-hour cache TTL instead of the default five minutes
cache_tools	bool	False	Add cache control to the last tool definition
system_prompt_blocks	Optional[Union[List[SystemPromptBlock], Callable[[], List[SystemPromptBlock]]]]	None	System-prompt blocks with per-block cache controls. A callable is evaluated for each request.
request_params	Optional[Dict[str, Any]]	None	Additional Anthropic request parameters. These are applied after the dedicated fields above.
betas	Optional[List[str]]	None	Exact Anthropic beta names sent through the SDK’s beta Messages API
context_management	Optional[Dict[str, Any]]	None	Anthropic context-management request body. Add the beta required by the selected strategy to betas.
mcp_servers	Optional[List[MCPServerConfiguration]]	None	Remote MCP server connections. The current connector also requires mcp-client-2025-04-04 in betas.
skills	Optional[List[Dict[str, str]]]	None	Agent Skills to load, such as [{"type": "anthropic", "skill_id": "pptx", "version": "latest"}]. Agno adds the Skills beta, the legacy code-execution beta, and the code_execution_20250825 tool.
citations	bool	True	Attach citations to URL, local, and bytes document blocks. Agno v2.7.2 omits citations from Anthropic file-ID blocks and suppresses them when structured output is active.
append_trailing_user_message	Optional[bool]	None	Append a user turn when a conversation ends with an assistant message. Agno enables this by default for Claude 4.6 and later models, which do not support assistant prefill.
trailing_user_message_content	str	"continue"	Content of the appended user message
api_key	Optional[str]	None	Anthropic API key. Falls back to ANTHROPIC_API_KEY.
auth_token	Optional[str]	None	Anthropic auth token. Falls back to ANTHROPIC_AUTH_TOKEN.
default_headers	Optional[Dict[str, Any]]	None	Default headers for every request
timeout	Optional[float]	None	Request timeout in seconds
http_client	Optional[Union[httpx.Client, httpx.AsyncClient]]	None	Custom synchronous or asynchronous HTTPX client
client_params	Optional[Dict[str, Any]]	None	Additional Anthropic client constructor parameters
client	Optional[AnthropicClient]	None	Preconfigured synchronous Anthropic client
async_client	Optional[AsyncAnthropicClient]	None	Preconfigured asynchronous Anthropic client

Claude is a subclass of the Model class and supports the inherited model parameters.