Skip to main content

AgentMark Pydantic AI Adapter

Pydantic AI adapter for AgentMark - integrate AgentMark prompts with Pydantic AI for type-safe LLM interactions in Python.

Installation

pip install agentmark-pydantic-ai-v0

For specific providers, install with extras:

pip install agentmark-pydantic-ai-v0[openai]
pip install agentmark-pydantic-ai-v0[anthropic]
pip install agentmark-pydantic-ai-v0[gemini]

Quick Start

from agentmark_pydantic_ai_v0 import create_pydantic_ai_client, run_text_prompt

# Create client
client = create_pydantic_ai_client()

# Load and format a prompt
prompt = await client.load_text_prompt(ast)  # AST from AgentMark compiler
params = await prompt.format(props={"name": "Alice"})

# Execute with runner utility
result = await run_text_prompt(params)
print(result.output)

# Or use Pydantic AI directly
from pydantic_ai import Agent

agent = Agent(params.model, system_prompt=params.system_prompt)
result = await agent.run(params.user_prompt)
print(result.output)

Features

  • Type-safe integration: Full type safety from AgentMark prompts to Pydantic AI
  • Model registry: Flexible model name resolution with pattern matching
  • Native tools: Pass pydantic-ai Tool objects or callables directly
  • MCP support: Model Context Protocol integration for external tool servers
  • Structured output: Automatic JSON Schema to Pydantic model conversion
  • Runner utilities: Convenience functions for common execution patterns

API Reference

Factory Function

from agentmark_pydantic_ai_v0 import create_pydantic_ai_client

client = create_pydantic_ai_client(
    model_registry=None,      # Optional custom model registry
    tools=None,               # Optional list of native Tool objects or callables
    mcp_registry=None,        # Optional MCP server registry
    eval_registry=None,       # Optional eval registry
    loader=None,              # Optional prompt loader
)

Model Registry

from agentmark_pydantic_ai_v0 import PydanticAIModelRegistry
import re

# Default registry (passthrough — model names forwarded as-is to Pydantic AI)
registry = PydanticAIModelRegistry.create_default()

# Register providers you need (user-driven, not pre-registered)
registry = (
    PydanticAIModelRegistry.create_default()
    .register_providers({"openai": "openai", "anthropic": "anthropic"})
)

# Or register specific models
registry = PydanticAIModelRegistry()
registry.register_models("gpt-4o", lambda name, opts: f"openai:{name}")
registry.register_models(
    re.compile(r"^claude-"),
    lambda name, opts: f"anthropic:{name}"
)

Tools

Tools are passed as native pydantic-ai Tool objects or plain callables:

from pydantic_ai import Tool

# Using callables (name is inferred from function name)
def search(query: str) -> str:
    return search_web(query)

client = create_pydantic_ai_client(tools=[search])

# Using Tool objects for more control
tool = Tool(function=search, name="search", description="Search the web")
client = create_pydantic_ai_client(tools=[tool])

The MDX config references tools by name. Only tools whose names match entries in the MDX tools list will be included at adapt time.

MCP Server Registry

from agentmark_pydantic_ai_v0 import McpServerRegistry

registry = McpServerRegistry()

# Register HTTP-based MCP server
registry.register("search-server", {
    "url": "http://localhost:8000/mcp",
    "headers": {"Authorization": "Bearer token"},  # Optional
})

# Register stdio-based MCP server
registry.register("python-runner", {
    "command": "python",
    "args": ["-m", "mcp_server"],
    "cwd": "/app",
    "env": {"API_KEY": "secret"},
})

# Use with client
client = create_pydantic_ai_client(mcp_registry=registry)

Then in your AgentMark prompt, reference MCP tools:

tools:
  - mcp://search-server/web-search    # Single MCP tool
  - mcp://search-server/*             # All tools from MCP server
  - search                            # Native tool by name

Runner Utilities

from agentmark_pydantic_ai_v0 import run_text_prompt, run_object_prompt, stream_text_prompt

# Run text prompt
result = await run_text_prompt(params)
print(result.output)       # str
print(result.usage)        # Usage stats

# Run object prompt (structured output)
result = await run_object_prompt(params)
print(result.output)       # Typed Pydantic model

# Stream text prompt
async for chunk in stream_text_prompt(params):
    print(chunk, end="", flush=True)

Webhook Handler

For building HTTP servers that execute AgentMark prompts (used by the CLI dev server):

from agentmark_pydantic_ai_v0 import create_pydantic_ai_client, PydanticAIWebhookHandler

# Create client and handler
client = create_pydantic_ai_client()
handler = PydanticAIWebhookHandler(client)

# Execute a prompt (non-streaming)
result = await handler.run_prompt(prompt_ast, {"shouldStream": False})
print(result["result"])  # "Hello, world!"

# Execute a prompt (streaming)
result = await handler.run_prompt(prompt_ast, {"shouldStream": True})
async for chunk in result["stream"]:
    print(chunk)  # NDJSON chunks

The webhook handler implements the AgentMark webhook protocol, producing NDJSON responses compatible with the CLI.

License

MIT

Metadata

Release files for agentmark-pydantic-ai-v0 0.5.0

For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.

Source distribution (sdist)

Source distribution for agentmark-pydantic-ai-v0 0.5.0
File Size Uploaded Provenance Actions
agentmark_pydantic_ai_v0-0.5.0.tar.gz 234.2 kB Without provenance Details for agentmark_pydantic_ai_v0-0.5.0.tar.gz

Built distribution (wheel)

Table of built distributions (wheels) for agentmark-pydantic-ai-v0 0.5.0
File Interpreter ABI Platform Actions
agentmark_pydantic_ai_v0-0.5.0-py3-none-any.whl Python 3 none any Details for agentmark_pydantic_ai_v0-0.5.0-py3-none-any.whl

Total release size: 260.6 kB

Release files / agentmark_pydantic_ai_v0-0.5.0.tar.gz

Download URL agentmark_pydantic_ai_v0-0.5.0.tar.gz
Size 234.2 kB
Tags Source
SHA-256 checksum
How to use checksums
611c80a8050ee3d5d139b25d33836cefc18c0f97db8ddb7fe86d27adc79a5ce5
BLAKE2b-256 checksum
How to use checksums
48f105c22088b140fc8de074c944fca8223086b2d9e978e3416d615624ccfb01
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
No
Uploaded via twine/6.2.0 CPython/3.12.13

Release files / agentmark_pydantic_ai_v0-0.5.0-py3-none-any.whl

Download URL agentmark_pydantic_ai_v0-0.5.0-py3-none-any.whl
Size 26.4 kB
Tags Python 3
SHA-256 checksum
How to use checksums
e57220051ff686e673df1ead370974f4d112ee83959895a3f8952edf00046e6b
BLAKE2b-256 checksum
How to use checksums
1591ec1e10e3779ff185ad40466f918227ab97448f90546cdbb05916496bc3a9
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
No
Uploaded via twine/6.2.0 CPython/3.12.13

Security for agentmark-pydantic-ai-v0 0.5.0

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

No provenance found

None of the files in this release were uploaded with provenance metadata.

A note on package security

PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.

To learn more about protecting yourself from malicious actors, visit our package security documentation.

Anthropic, PBC Visionary sponsor Bloomberg Visionary sponsor Hudson River Trading Visionary sponsor Meta Visionary sponsor NVIDIA Visionary sponsor Microsoft Sustainability sponsor Depot Continuous Integration AWS Cloud computing and Security Sponsor Datadog Monitoring Fastly CDN Google Download Analytics Sentry Error logging StatusPage Status page