Repository navigation
build: lift the litellm cap; 1.109 fixed the mcp import it was for - #687
Merged
Merged
Conversation
LiteLLM 1.106 through 1.108.3 import the proxy MCP handler on every call with tools, and that import fails under mcp<2.2 (BerriAI/litellm#46045). 1.108.4 and 1.109.0 onward import it only for a gateway tool (#45961). The floor is now 1.109.0 with no upper bound: 1.106.*, 1.107.* (1.107.0 was malware) and 1.108.0-1.108.3 all sit below it. Fresh installs move to openai 3 and openai-agents 0.25. Kept for an install below the floor (another package can pin litellm lower, and pip lets the last install win): _drop_repeated_thinking, and the Claude "none" ladder and Databricks fill, which 1.110 makes redundant but 1.109 still needs. Tests: - The tripwires asserting LiteLLM still repeats streamed thinking, still sends Claude no thinking for "none", and still drops a Databricks switch beside "none" are removed: 1.109 or 1.110 fixed each, so they fail on every version CI installs. The behaviour tests of the kept code stay. - No LiteLLM in range repeats streamed thinking any more, so _drop_repeated_thinking gets a direct test with both stream shapes. - openai-agents 0.25 adds ModelSettings.timeout; LitellmModel does not forward it, so only the samples change. - From 1.109 LiteLLM passes an unknown Bedrock model's reasoning_effort through instead of dropping it, so the summary ladder tries it. Verified: a real gpt-6-luna chat with tools under mcp 1.30.0 works on litellm 1.109.0 and fails on 1.106.10 with the mcp hint. Full suite on litellm 1.109.0 and 1.110.0: py3.13 with frameworks 1270 passed, 7 skipped; py3.10 without 1097 passed, 180 skipped.
Codex Review Summary: ✅ CompletedThis comment shows the latest Codex review activity on this pull request.
ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings. |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Installs get current LiteLLM again instead of staying on 1.105, and own-model chat keeps working with an older mcp: the
<1.106cap gives way tolitellm>=1.109.0.LiteLLM 1.106 through 1.108.3 import the proxy MCP handler on every call with tools, and that import fails under mcp<2.2 (BerriAI/litellm#46045). 1.108.4 and 1.109.0 onward import it only for a gateway tool (#45961). Every broken release, 1.107.0 included, sits below the new floor, so no
!=list and no upper bound are needed. Fresh installs move to openai 3 and openai-agents 0.25.Changes
pyproject.toml,requirements.txt:litellm>=1.109.0, cap and its comments gone._drop_repeated_thinking, and the Claude "none" ladder and Databricks fill, which 1.110 makes redundant but 1.109 still needs. Two comments now say which versions need them._drop_repeated_thinkingwith both stream shapes, since no LiteLLM in range produces the repeat any more.ModelSettings.timeout(openai-agents 0.25) added to the settings samples;LitellmModeldoes not forward it.reasoning_effortthrough instead of dropping it, so the summary ladder tries it; the test expects that now.Verification
gpt-6-lunachat with tools under mcp 1.30.0 works on litellm 1.109.0 and fails on 1.106.10 with the mcp hint._drop_repeated_thinking(never blanks; blanks any closing text).