Fix mcp large response race condition #5065
Merged
Conversation
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
alexyao2015
force-pushed
the
fix/mcp-response-race
branch
2 times, most recently
from
6feb213 to
839250f
Compare
When multiple subagents are running in parallel and they call an mcp tool that returns a large response, these responses may all be returned at the exact same second which can result in subagents receiving paths containing the incorrect information. We fix this by adding microseconds to the filename so that this is very unlikely to occur anymore. Maybe a better solution is to check if the file exists, but I think this is sufficient for now. Signed-off-by: alexyao2015 <[email protected]>
alexyao2015
force-pushed
the
fix/mcp-response-race
branch
from
839250f to
539977d
Compare
DOsinga
reviewed
Oct 8, 2025
Collaborator
DOsinga
left a comment
DOsinga
left a comment
There was a problem hiding this comment.
this is for debugging only so I think the current behavior is fine. if anything we should just delete the file
Contributor
Author
|
This is not just for debugging. If an mcp server provides a large response to the agent, it's stored in a file which can be erroneously overwritten with incorrect content in the current implementation. |
Collaborator
|
well, that is for debugging yes. to be able to see what the large response was. I think this is something we used to have trouble with and can probably go /cc @michaelneale |
Contributor
Author
|
If by debugging, you mean the alternative is using an in memory temporary file, I suppose so? This file is currently read by the agent using shell commands if the developer tools extension is enabled, so it's definitely not used only for debugging as you might think it is. With the race condition existing as it is today, sub agents get confused because they are fed the wrong data when they receive a response from the mcp server. |
Collaborator
|
ok, fine. let's give it a real unique name though instead of still a timestmap |
DOsinga
approved these changes
Oct 11, 2025
Collaborator
|
yes, this makes sense, and wow, this is a wild edge case. Yeah some MCPs will spew a lot, so this directs to a file, if they happen to have same name... ouch. It would rarely happen as responses now have to be pretty large (if we make it too small, it slows down common MCP calls). Belatedly approve this @alexyao2015 wow, that is cool fix. |
michaelneale
added a commit
that referenced
this pull request
Oct 16, 2025
* main: Skip hidden & real format (#5194) docs: Hacktoberfest blog submission - Best Practices for Using Goose in Enterprise Environments by Anudhyan Datta. (#5184) docs: prompt injection detection (#5193) Fix mcp large response race condition (#5065) Compaction overhaul (#5186) fix: #3960 better approach to input schema for dynamic task params (#5189) used recipe id or deeplink to start agent (#5154) [docs] Add Blog Post: "Designing AI for Users, Not Just LLMs" (#5190) docs: update cognee, jetbrains, mbot extensions config (#5172) Minimally disable subagents when not in autonomous model (#5149) Fix provider sort (#5188) blog: Getting Started with Goose on Windows (#5156) feat: add CI/CD Pipeline recipe (#5183) feat: add Daily Standup Report Generator recipe (#5123) (#5131) Sort providers in alphabetical vs random (#5090) Declarative providers (#5084) adding youtube link to firecrawl mcp tutorial, merge after 9am Eastern Oct 15 (#5173) Ollama integration: modified default model + added models (#5153) Fix codex subagent configuration in documentation (#5180)
wpfleger96
added a commit
to wpfleger96/goose
that referenced
this pull request
Oct 16, 2025
* main: Blog: Best Practices for Prompt Engineering with goose (block#5204) force WAL sync after session create (block#5202) Feat: goose Apify MCP integration docs (block#5047) feat: enhance goose to search sessions for easy recall (block#5177) Skip hidden & real format (block#5194) docs: Hacktoberfest blog submission - Best Practices for Using Goose in Enterprise Environments by Anudhyan Datta. (block#5184) docs: prompt injection detection (block#5193) Fix mcp large response race condition (block#5065)
Merged
michaelneale
added a commit
that referenced
this pull request
Oct 17, 2025
* main: (119 commits) Break compaction back into check_ and do_ compaction (#5212) fix: revert built app name to uppercase Goose (#5206) feat: add Code Documentation Generator recipe (#5121) (#5125) Revert "feat: enhance goose to search sessions for easy recall (#5177)" (#5209) Blog: Best Practices for Prompt Engineering with goose (#5204) force WAL sync after session create (#5202) Feat: goose Apify MCP integration docs (#5047) feat: enhance goose to search sessions for easy recall (#5177) Skip hidden & real format (#5194) docs: Hacktoberfest blog submission - Best Practices for Using Goose in Enterprise Environments by Anudhyan Datta. (#5184) docs: prompt injection detection (#5193) Fix mcp large response race condition (#5065) Compaction overhaul (#5186) fix: #3960 better approach to input schema for dynamic task params (#5189) used recipe id or deeplink to start agent (#5154) [docs] Add Blog Post: "Designing AI for Users, Not Just LLMs" (#5190) docs: update cognee, jetbrains, mbot extensions config (#5172) Minimally disable subagents when not in autonomous model (#5149) Fix provider sort (#5188) blog: Getting Started with Goose on Windows (#5156) ...
michaelneale
added a commit
that referenced
this pull request
Oct 17, 2025
* main: (143 commits) Break compaction back into check_ and do_ compaction (#5212) fix: revert built app name to uppercase Goose (#5206) feat: add Code Documentation Generator recipe (#5121) (#5125) Revert "feat: enhance goose to search sessions for easy recall (#5177)" (#5209) Blog: Best Practices for Prompt Engineering with goose (#5204) force WAL sync after session create (#5202) Feat: goose Apify MCP integration docs (#5047) feat: enhance goose to search sessions for easy recall (#5177) Skip hidden & real format (#5194) docs: Hacktoberfest blog submission - Best Practices for Using Goose in Enterprise Environments by Anudhyan Datta. (#5184) docs: prompt injection detection (#5193) Fix mcp large response race condition (#5065) Compaction overhaul (#5186) fix: #3960 better approach to input schema for dynamic task params (#5189) used recipe id or deeplink to start agent (#5154) [docs] Add Blog Post: "Designing AI for Users, Not Just LLMs" (#5190) docs: update cognee, jetbrains, mbot extensions config (#5172) Minimally disable subagents when not in autonomous model (#5149) Fix provider sort (#5188) blog: Getting Started with Goose on Windows (#5156) ...
zanesq
added a commit
that referenced
this pull request
Oct 17, 2025
* 'main' of github.com:block/goose: (22 commits) Rewrite extension management tools (#5057) fix: re-sync package-lock.json (#5235) docs: Hacktoberfest MCP youtube short entry to community-content.json (#5150) feat: add schedule button to recipe entries (#5217) Autocompact threshold UI cleanup (#5232) fix: correct schema for openai tools (#5229) Break compaction back into check_ and do_ compaction (#5212) fix: revert built app name to uppercase Goose (#5206) feat: add Code Documentation Generator recipe (#5121) (#5125) Revert "feat: enhance goose to search sessions for easy recall (#5177)" (#5209) Blog: Best Practices for Prompt Engineering with goose (#5204) force WAL sync after session create (#5202) Feat: goose Apify MCP integration docs (#5047) feat: enhance goose to search sessions for easy recall (#5177) Skip hidden & real format (#5194) docs: Hacktoberfest blog submission - Best Practices for Using Goose in Enterprise Environments by Anudhyan Datta. (#5184) docs: prompt injection detection (#5193) Fix mcp large response race condition (#5065) Compaction overhaul (#5186) fix: #3960 better approach to input schema for dynamic task params (#5189) ...
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
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.
Summary
When multiple subagents are running in parallel and they call an mcp
tool that returns a large response, these responses may all be returned
at the exact same second which can result in subagents receiving paths
containing the incorrect information.
We fix this by adding microseconds to the filename so that this is very
unlikely to occur anymore. Maybe a better solution is to check if the
file exists, but I think this is sufficient for now.
Type of Change
Testing
Tested that this race condition no longer occurs when the main agent calls subagents to run in parallel.