Skip to content

LiteLLM 1.107.0-rc.2 previews default-deny agent access and MCP updates

Updated LiteLLM

Summary

A release candidate adds default-deny agent access, verified stateless MCP capabilities, and token-counting changes; it is not a stable release.

Category
Agent infra
Date
Pricing
Not stated in the sources we cited.

v1.107.0-rc.2 was published at 2026-10-20T03:22:42Z. This summary follows the maintainers’ release notes.

What changed

  • The proxy adds agent_access_default_deny for least-privilege agent access.
  • MCP changes enable verified stateless protocol capabilities and backport the ID-JAG stack.
  • OAuth consent can be scoped to x-mcp-servers with an explicit scope setting, and Google upstream OAuth requests offline access.
  • The candidate backports a count_tokens stack flagged as breaking. It also lets Gemini callers disable thought summaries while keeping thinking enabled.

Before upgrading

GitHub marks this release as a prerelease. Review the breaking token-counting and Terraform additional_drop_params changes, and test actual provider routes before a production migration. This candidate is distinct from the previously covered 1.107.0-dev.3, not a newly announced stable 1.107.0.

Explore the litellm hub and agent-infra category.

More from LiteLLM

View all

Questions

What is LiteLLM?

According to its docs, LiteLLM is an open-source library that gives you a single, unified interface to call 100+ LLMs using the OpenAI format.

Is LiteLLM open source?

The docs describe it as an open-source library, but the GitHub API does not report a recognized license for the repository.

Filed on Agent Radar. Claims in this note are limited to the sources above.