fix(code): serialize MCP OAuth token refreshes to prevent reuse revocation - #4565
Merged
Conversation
…ation Two dcode processes or provider instances sharing an MCP token file could concurrently replay the same refresh token. The LangSmith OAuth server rotates refresh tokens and revokes the whole identity+client token family on reuse, so this surfaced as requests hanging until a full re-auth. Guard the refresh with a cross-process `filelock` keyed on a sibling lock file, reload tokens from disk after acquiring it (so a peer's rotation is observed and the refresh is skipped when the token is already valid), and only then perform and persist the SDK refresh grant. Blocking file IO runs in a worker thread to keep the server event loop responsive. Co-authored-by: open-swe[bot] <open-swe@users.noreply.github.com>
github-actions
Bot
added
dcode
Mason Daugherty (mdrxy)
marked this pull request as ready for review
July 8, 2026 20:21
![open-swe[bot]](https://avatars.githubusercontent.com/in/1395372?s=60&v=4){kind=small}
Mason Daugherty (mdrxy)
added a commit
that referenced
this pull request
Jul 9, 2026
> [!CAUTION] > Merging this PR will automatically publish to **PyPI** and create a **GitHub release**. For the full release process, see [`.github/RELEASING.md`](https://github.com/langchain-ai/deepagents/blob/main/.github/RELEASING.md). --- _Everything below this line will be the GitHub release body._ --- ## [0.1.35](deepagents-code==0.1.34...deepagents-code==0.1.35) (2026-07-09) ### Features * Restore interrupted prompt to input on ESC ([#4544](#4544)) ([fccf037](fccf037)) * Add `[startup].mode` default approval mode ([#4573](#4573)) ([7c5bf54](7c5bf54)) * Offer restart after saving Tavily key via `/auth` ([#4560](#4560)) ([12df81a](12df81a)) * Reload env from `/auth` modal via Ctrl+R ([#4566](#4566)) ([f07d638](f07d638)) * Toast on saved `/auth` API key ([#4558](#4558)) ([ee3c264](ee3c264)) ### Bug Fixes * Harden approval content rendering ([#4581](#4581)) ([38446fd](38446fd)) * Preserve transcript order during virtualization ([#4549](#4549)) ([f6ee70c](f6ee70c)) * Run stdio MCP server pre-flight check off the event loop ([#4434](#4434)) ([c9636e2](c9636e2)) * Avoid duplicate "criteria ready" message on `/goal` revisions ([#4559](#4559)) ([1110497](1110497)) * Restore welcome banner tips ([#4528](#4528)) ([3f1e55e](3f1e55e)) * Clarify managed `rg` install failures ([#4578](#4578)) ([434c84a](434c84a)) * Dedupe update/install log path output ([#4553](#4553)) ([1398fee](1398fee)) * Keep notification center open for API-key entry ([#4568](#4568)) ([6e89417](6e89417)) * Queue `/mcp login` sent before the server connects ([#4533](#4533)) ([edac82c](edac82c)) * Serialize MCP OAuth token refreshes to prevent reuse revocation ([#4565](#4565)) ([c37100d](c37100d)) --- _Everything above this line will be the GitHub release body._ --- > [!NOTE] > A **New Contributors** section is appended to the GitHub release notes automatically at publish time (see [Release Pipeline](https://github.com/langchain-ai/deepagents/blob/main/.github/RELEASING.md#release-pipeline), step 2). --------- Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com> Co-authored-by: Mason Daugherty <github@mdrxy.com>
Marcelo5444
pushed a commit
to Marcelo5444/deepagents
that referenced
this pull request
Jul 30, 2026
…ation (langchain-ai#4565) Fixed frequent LangSmith MCP OAuth timeouts/re-auth prompts caused by concurrent token refreshes invalidating the session; refreshes are now serialized across dcode processes and provider instances. --- dcode stores MCP OAuth tokens per-server under `~/.deepagents/.state/mcp-tokens/`. The MCP SDK's `OAuthClientProvider` serializes refreshes with an in-memory `OAuthContext.lock`, which only covers a single provider instance. Two dcode processes — or two provider instances in one process (e.g. a throwaway discovery session created during tool loading alongside the cached runtime session) — can each read the same refresh token from disk and `POST` the `refresh_token` grant concurrently. The LangSmith OAuth server (in `langchainplus`) rotates refresh tokens and, on replay of an already-rotated token, treats it as reuse and revokes every active refresh token for that identity+client. That is why the LangSmith MCP worked for a few minutes (one access-token lifetime) and then hung until a full re-auth. This adds a cross-process lock (`filelock`) keyed on a sibling `.lock` file next to each token file. Before refreshing, `_ExpiryAwareOAuthClientProvider` acquires the lock, reloads tokens from disk (so a peer's rotation is observed and the refresh is skipped when the token is already valid), and only then performs and persists the SDK refresh grant. The lock wait is bounded (falls back to a best-effort refresh on timeout so a crashed peer can't hang tool calls), and all blocking file IO runs in a worker thread to keep the `blockbuster`-guarded server loop responsive. The lock file holds no token material. `filelock` is added as a direct dependency (`>=3.12,<4.0.0`); it was already resolved transitively (via `huggingface-hub`) at `3.29.4`. It is MIT-licensed and actively maintained. Made by [Open SWE](https://openswe.vercel.app/agents/2c15284a-7a9d-8076-66e9-3c97ab30976c) --------- Co-authored-by: open-swe[bot] <open-swe@users.noreply.github.com>
Marcelo5444
pushed a commit
to Marcelo5444/deepagents
that referenced
this pull request
Jul 30, 2026
> [!CAUTION] > Merging this PR will automatically publish to **PyPI** and create a **GitHub release**. For the full release process, see [`.github/RELEASING.md`](https://github.com/langchain-ai/deepagents/blob/main/.github/RELEASING.md). --- _Everything below this line will be the GitHub release body._ --- ## [0.1.35](langchain-ai/deepagents@deepagents-code==0.1.34...deepagents-code==0.1.35) (2026-07-09) ### Features * Restore interrupted prompt to input on ESC ([langchain-ai#4544](langchain-ai#4544)) ([fccf037](langchain-ai@fccf037)) * Add `[startup].mode` default approval mode ([langchain-ai#4573](langchain-ai#4573)) ([7c5bf54](langchain-ai@7c5bf54)) * Offer restart after saving Tavily key via `/auth` ([langchain-ai#4560](langchain-ai#4560)) ([12df81a](langchain-ai@12df81a)) * Reload env from `/auth` modal via Ctrl+R ([langchain-ai#4566](langchain-ai#4566)) ([f07d638](langchain-ai@f07d638)) * Toast on saved `/auth` API key ([langchain-ai#4558](langchain-ai#4558)) ([ee3c264](langchain-ai@ee3c264)) ### Bug Fixes * Harden approval content rendering ([langchain-ai#4581](langchain-ai#4581)) ([38446fd](langchain-ai@38446fd)) * Preserve transcript order during virtualization ([langchain-ai#4549](langchain-ai#4549)) ([f6ee70c](langchain-ai@f6ee70c)) * Run stdio MCP server pre-flight check off the event loop ([langchain-ai#4434](langchain-ai#4434)) ([c9636e2](langchain-ai@c9636e2)) * Avoid duplicate "criteria ready" message on `/goal` revisions ([langchain-ai#4559](langchain-ai#4559)) ([1110497](langchain-ai@1110497)) * Restore welcome banner tips ([langchain-ai#4528](langchain-ai#4528)) ([3f1e55e](langchain-ai@3f1e55e)) * Clarify managed `rg` install failures ([langchain-ai#4578](langchain-ai#4578)) ([434c84a](langchain-ai@434c84a)) * Dedupe update/install log path output ([langchain-ai#4553](langchain-ai#4553)) ([1398fee](langchain-ai@1398fee)) * Keep notification center open for API-key entry ([langchain-ai#4568](langchain-ai#4568)) ([6e89417](langchain-ai@6e89417)) * Queue `/mcp login` sent before the server connects ([langchain-ai#4533](langchain-ai#4533)) ([edac82c](langchain-ai@edac82c)) * Serialize MCP OAuth token refreshes to prevent reuse revocation ([langchain-ai#4565](langchain-ai#4565)) ([c37100d](langchain-ai@c37100d)) --- _Everything above this line will be the GitHub release body._ --- > [!NOTE] > A **New Contributors** section is appended to the GitHub release notes automatically at publish time (see [Release Pipeline](https://github.com/langchain-ai/deepagents/blob/main/.github/RELEASING.md#release-pipeline), step 2). --------- Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com> Co-authored-by: Mason Daugherty <github@mdrxy.com>
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
1 participant
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.
Fixed frequent LangSmith MCP OAuth timeouts/re-auth prompts caused by concurrent token refreshes invalidating the session; refreshes are now serialized across dcode processes and provider instances.
dcode stores MCP OAuth tokens per-server under
~/.deepagents/.state/mcp-tokens/. The MCP SDK'sOAuthClientProviderserializes refreshes with an in-memoryOAuthContext.lock, which only covers a single provider instance. Two dcode processes — or two provider instances in one process (e.g. a throwaway discovery session created during tool loading alongside the cached runtime session) — can each read the same refresh token from disk andPOSTtherefresh_tokengrant concurrently.The LangSmith OAuth server (in
langchainplus) rotates refresh tokens and, on replay of an already-rotated token, treats it as reuse and revokes every active refresh token for that identity+client. That is why the LangSmith MCP worked for a few minutes (one access-token lifetime) and then hung until a full re-auth.This adds a cross-process lock (
filelock) keyed on a sibling.lockfile next to each token file. Before refreshing,_ExpiryAwareOAuthClientProvideracquires the lock, reloads tokens from disk (so a peer's rotation is observed and the refresh is skipped when the token is already valid), and only then performs and persists the SDK refresh grant. The lock wait is bounded (falls back to a best-effort refresh on timeout so a crashed peer can't hang tool calls), and all blocking file IO runs in a worker thread to keep theblockbuster-guarded server loop responsive. The lock file holds no token material.filelockis added as a direct dependency (>=3.12,<4.0.0); it was already resolved transitively (viahuggingface-hub) at3.29.4. It is MIT-licensed and actively maintained.Made by Open SWE