Skip to content

SDK & MCP v1.3.0: Catalog, Governance, Identity, and Run Operations

RunAgents v1.3.0 brings the public Python SDK and MCP server much closer to the management surface already available in the CLI.

This release matters if you use:

  • Python automation against RunAgents APIs
  • Claude Code, Cursor, Codex, or similar assistants through runagents-mcp
  • the public SDK as the foundation for internal deployment and operations workflows

Highlights

  • Catalog parity — assistants and Python automation can now list catalog agents, inspect manifests, list versions, and deploy directly from catalog blueprints.
  • Governance parity — policies and approval connectors are now first-class SDK and MCP resources instead of CLI-only workflows.
  • Identity parity — identity providers can now be listed, inspected, created, updated, and deleted from Python and MCP.
  • Run operations parity — the SDK and MCP server now support richer run inspection, including get, event filtering, timelines, waiting for terminal state, and full run export.
  • Deploy and approval parity — deploy flows now accept policies, identity providers, drafts, artifacts, and framework hints; approvals now expose scoped approval parameters.

What shipped

Python SDK

The Python client now exposes:

  • client.catalog
  • client.policies
  • client.approval_connectors
  • client.identity_providers

It also expands:

  • client.runs
  • list
  • get
  • events
  • timeline
  • wait
  • export
  • client.agents.deploy(...)
  • source deploy options
  • draft and artifact deploy paths
  • policy binding
  • identity-provider binding
  • client.approvals.approve(...)
  • scope
  • duration

MCP server

The MCP server now supports assistant-facing tools for:

  • catalog discovery and deploy
  • policy read/apply/delete/translate
  • approval connector configuration, defaults, testing, and activity
  • identity provider management
  • richer run debugging and export
  • scoped approval decisions

Why this release exists

Before v1.3.0, the CLI had moved ahead of the public Python SDK and MCP story.

That meant the public repo could truthfully say RunAgents supported:

  • catalog deployment
  • policy workflows
  • approval connectors
  • identity providers
  • richer run operations

but not all of those were available consistently from:

  • the Python SDK
  • the MCP server used by coding assistants

v1.3.0 closes much of that gap and makes the public SDK/MCP surface feel more like a mature product ecosystem rather than a narrow helper package.

Release guidance

Move to v1.3.0 if you want:

  • assistant workflows that can inspect and manage more of the real platform
  • Python automation for catalog, governance, identity, and operations
  • a more consistent public story across CLI, SDK, MCP, docs, and releases

Notes

  • The MCP command still ships through pip install runagents[mcp].
  • This release does not introduce a separately published runagents-mcp package.
  • The CLI release channels remain unchanged:
  • GitHub releases
  • npm
  • Homebrew
  • S3 install script
  • The Python SDK and MCP surface ship through the unified versioned Python package on PyPI.