Pydantic AI

OrchestrationMIT

Pydantic AI is a Python SDK for building agents with a typed agent loop. Developers define tools as functions whose arguments are validated before the code runs, request structured outputs, and inject typed dependencies. It connects to model providers including OpenAI, Anthropic, Google, Bedrock and Ollama, can use MCP servers as tools, and emits OpenTelemetry instrumentation. For long-running work the documentation lists nine durable execution engines, among them Temporal, DBOS and Prefect. It installs with pip or uv, and a separate Pydantic Evals package grades agent outputs.

The repository README also lists realtime voice sessions, image generation, a terminal coding agent and UI event streams such as AG-UI and Vercel AI.

Pinout

  • MCP server: supported, verified
  • HTTP API: not checked yet
  • command line: supported, verified
  • Python SDK: supported, verified
  • TypeScript SDK: not checked yet
  • OpenAPI description: not checked yet
Where each pin comes from
InterfaceStatusSource we readRead on
MCP serverSupported, verifiedDocumentation, MCP overview (agents connect to MCP servers)
HTTP APINot checked yet——
Command lineSupported, verifiedGitHub README, interfaces list (CLI)
Python SDKSupported, verifiedDocumentation overview, install (pip install pydantic-ai)
TypeScript SDKNot checked yet——
OpenAPI descriptionNot checked yet——

Facts

Facts about Pydantic AI
LicenseMIT. GitHub repository page, README, read .
Model providersOpenAI, Anthropic, Google, Bedrock, Azure AI Foundry, Groq, Mistral, xAI and Ollama named, plus dozens more. Documentation overview, read .
TypingTyped end to end: structured outputs, typed dependency injection and typed tools. Documentation overview, read .
Durable executionNine engines listed: Temporal, DBOS, Prefect, Restate, AWS Lambda, Airflow, Absurd, pgtask and Azure Durable Task. Documentation overview, read .
MCP clientAgents can connect to MCP servers and use their tools; the MCP capability runs the server locally by default. Documentation, MCP overview, read .
MCP server sideThe documentation states agents can be used within MCP servers. Documentation, MCP overview, read .
ObservabilityOpenTelemetry-native instrumentation emitting spans for model and tool calls; any OTLP backend works. Documentation overview, read .
EvalsPydantic Evals grades an agent's final outputs, including tool-call trajectories, on datasets or live production traffic. Documentation overview, read .
Latest releasev2.68.0, dated 2026-11-11 in the release heading. GitHub Releases page, read .
InterfacesCLI, built-in web chat, realtime speech, UI event streams (AG-UI, Vercel AI), ACP and GitHub Agentic Workflows. GitHub repository page, README, read .
Last commit. GitHub API, latest commit on the default branch of pydantic/pydantic-ai, read .

Call example

uv add pydantic-ai

Limits

  • The documentation moved to pydantic.dev/docs/ai; ai.pydantic.dev redirects there.

Strengths

  • The library is MIT-licensed.
  • Nine durable execution engines are listed for long-running agents.
  • OpenTelemetry-native instrumentation works with any OTLP backend.

Alternatives to Pydantic AI

All orchestration tools for agents
  1. Durable execution platform for workflows, background jobs and AI agents.

    Orchestration
  2. Low-level framework for stateful, long-running agent workflows built as graphs.

    Orchestration
  3. Durable execution platform, open source, used to run long-lived agents.

    Orchestration
  4. Open-source Python framework for multi-agent workflows; separate commercial platform.

    Orchestration

Last check: . Method: public repository and docs read automatically, page and day written down. Next check due within 90 days. How we check

Is this your tool? Write to hello@agent-tools.directory from an address on its domain to correct a pin or a fact.