Prefect

What it is

An open-source, Python-native workflow orchestration engine: decorate functions as flows and tasks, then schedule, retry, observe and trigger them from events. The company also maintains FastMCP (Python framework for MCP servers) and sells Prefect Horizon, an enterprise MCP server platform.

Maker, ownership and history

Prefect (the company, remote-first) ships the open-source repo created 2018-06 (GitHub); Jeremiah Lowin is founder and CEO, and Prefect says it was founded in 2018 and has been “a profitable, fast-growing business” over the past year (company announcement of 2026-07-13, a self-description). Prefect 3.0 shipped in 2024 with open-sourced event-driven features. 2026-07-13: Prefect announced it is acquiring Dagster Labs; Dagster and Dagster+ keep their names, licences and contracts, and Prefect’s customer FAQ says no Prefect customer is asked to migrate and no roadmap changes are announced yet (dagster). Prefect describes three open-source product families (Prefect, Dagster, FastMCP): Dagster defines outcomes, Prefect executes the work, FastMCP governs access. Per dagster.io, the combined company is expected to operate under the Prefect name from August 2026 after closing. Funding history is not sourced.

Editions and deployment

Open source (Apache-2.0; latest 3.8.8 on 2026-10-06, PyPI) and Prefect Cloud plans Hobby (free), Starter, Team, Enterprise (SSO, SCIM, audit logging at higher tiers; managed compute minutes vs bring-your-own compute). Hybrid execution: work pools run on processes, containers, Kubernetes or cloud services while the control plane is hosted. Pricing: https://www.prefect.io/pricing

Core architecture

Flows/tasks as Python, state tracking with resume from last success, deployments, work pools, events and automations (reactive triggers), dynamic task creation at runtime.

Role in an enterprise AI rollout

Good fit for ML and agent workloads with dynamic, event-triggered runs (the Dagster letter describes Prefect as strong on dynamic workflows and ML teams). Horizon governs the MCP servers agents call. See data-pipelines-for-ai.

AI features as of October 2026

• Prefect MCP server for read-only diagnostics and docs access from assistants such as Claude (docs, fetched 2026-10-07).
• Prefect Horizon: Deploy, Registry, Gateway (tool-level RBAC, auth, audit), Remix; OAuth 2.1; 65+ identity providers; appears generally available (marked “NEW”). FastMCP claimed to power 70% of MCP servers (vendor claim).

Integrations

Kubernetes, Docker, AWS/GCP/Azure, dbt, Snowflake, Databricks; peers apache-airflow, dagster, kestra, temporal.

Strengths and weaknesses (opinion)

Strong: lowest-friction Python experience, dynamic workflows, MCP-governance angle. Weak: weaker asset lineage than Dagster; smaller ecosystem than Airflow.

Self-learning

• Docs: https://docs.prefect.io/ (free)
• FastMCP repo: https://github.com/PrefectHQ/fastmcp
• Prefect blog: https://www.prefect.io/blog

Sources (fetched 2026-10-07)

• https://docs.prefect.io/v3/get-started ; https://www.prefect.io/horizon ; https://www.prefect.io/pricing ; https://www.prefect.io/company
• https://dagster.io/blog/prefect-is-acquiring-dagster ; PyPI/GitHub API
• Second pass (2026-10-08): https://www.prefect.io/prefect-acquires-dagster ; https://www.prefect.io/prefect-dagster-faq ; https://dagster.io/prefect ; https://www.prefect.io/sitemap.xml (no funding or close announcement listed)

Open items

• Prefect funding history, acquisition terms and close date; Horizon GA date; whether Dagster and Prefect merge products (the FAQ announces no product changes).