[PR #28606] feat:add mcp session manager #32111

Closed
opened 2026-02-21 20:50:46 -05:00 by yindo · 0 comments
Owner

Original Pull Request: https://github.com/langgenius/dify/pull/28606

State: closed
Merged: No


Important

  1. Make sure you have read our contribution guidelines
  2. Ensure there is an associated issue and you have been assigned to it
  3. Use the correct syntax to link this PR: Fixes #<issue number>.

Summary

Fixes: https://github.com/langgenius/dify/issues/24224

Background

  • MCP tools opened/closed MCPClient per node, deleting sessions each time. Browser state (pages, cookies, etc.) vanished and every step paid full init, breaking multi-step flows like browser_navigate → browser_take_screenshot.

Fix

  • Cache MCPClient per workflow_execution_id via McpSessionManager/McpSessionRegistry.
  • Propagate execution ID through ToolRuntime → MCPTool → GraphEngine.
  • Auto-cleanup sessions when the workflow finishes.

Impact

  • Workflows reuse a single MCP session, preserving browser state and cutting init cost.
  • Measured: BROWSER_NAVIGATE ~63.3s → ~3.1s, BROWSER_TAKE_SCREENSHOT ~61.9s → ~0.17s.

Note

  • Agent Chat App currently stays on ephemeral MCP connections (no workflow_execution_id), so session reuse is not applied. Needs a separate issue to add an agent-specific scope ID and cleanup.

Tests

  • make lint and make type-check run at repo root (both succeeded with no errors).
  • uv run --project api --dev dev/pytest/pytest_unit_tests.sh (1,931 unit tests passed; only existing DeprecationWarnings).
  • End-to-end workflow run against Playwright MCP verifying BROWSER_INSTALL → BROWSER_NAVIGATE → BROWSER_TAKE_SCREENSHOT succeeds within a single session; confirmed runtime drop from ~63.3 s / ~61.9 s to ~3.1 s / ~0.17 s. Agent Chat mode exercised to ensure MCP calls still behave the same as before (per-call sessions, no regressions observed).

Screenshots

Before After
image image

Checklist

  • This change requires a documentation update, included: Dify Document
  • I understand that this PR may be closed in case there was no previous discussion or issues. (This doesn't apply to typos!)
  • I've added a test for each change that was introduced, and I tried as much as possible to make a single atomic change.
  • I've updated the documentation accordingly.
  • I ran dev/reformat(backend) and cd web && npx lint-staged(frontend) to appease the lint gods
**Original Pull Request:** https://github.com/langgenius/dify/pull/28606 **State:** closed **Merged:** No --- > [!IMPORTANT] > > 1. Make sure you have read our [contribution guidelines](https://github.com/langgenius/dify/blob/main/CONTRIBUTING.md) > 1. Ensure there is an associated issue and you have been assigned to it > 1. Use the correct syntax to link this PR: `Fixes #<issue number>`. ## Summary Fixes: https://github.com/langgenius/dify/issues/24224 <!-- Please include a summary of the change and which issue is fixed. Please also include relevant motivation and context. List any dependencies that are required for this change. --> ### Background - MCP tools opened/closed `MCPClient` per node, deleting sessions each time. Browser state (pages, cookies, etc.) vanished and every step paid full init, breaking multi-step flows like `browser_navigate → browser_take_screenshot`. ### Fix - Cache `MCPClient` per `workflow_execution_id` via `McpSessionManager/McpSessionRegistry`. - Propagate execution ID through `ToolRuntime → MCPTool → GraphEngine`. - Auto-cleanup sessions when the workflow finishes. ### Impact - Workflows reuse a single MCP session, preserving browser state and cutting init cost. - Measured: `BROWSER_NAVIGATE` ~63.3s → ~3.1s, `BROWSER_TAKE_SCREENSHOT` ~61.9s → ~0.17s. ### Note - Agent Chat App currently stays on ephemeral MCP connections (no `workflow_execution_id`), so session reuse is not applied. Needs a separate issue to add an agent-specific scope ID and cleanup. ### Tests - `make lint` and `make type-check` run at repo root (both succeeded with no errors). - `uv run --project api --dev dev/pytest/pytest_unit_tests.sh` (1,931 unit tests passed; only existing DeprecationWarnings). - End-to-end workflow run against Playwright MCP verifying BROWSER_INSTALL → BROWSER_NAVIGATE → BROWSER_TAKE_SCREENSHOT succeeds within a single session; confirmed runtime drop from ~63.3 s / ~61.9 s to ~3.1 s / ~0.17 s. Agent Chat mode exercised to ensure MCP calls still behave the same as before (per-call sessions, no regressions observed). ## Screenshots | Before | After | |--------|-------| | <img width="776" height="538" alt="image" src="https://github.com/user-attachments/assets/aa3e21ef-dd0e-4e66-a8e3-bc0f8e00f835" /> | <img width="776" height="538" alt="image" src="https://github.com/user-attachments/assets/91c7d2c2-1598-445c-948c-8c675ffbe653" /> | ## Checklist - [ ] This change requires a documentation update, included: [Dify Document](https://github.com/langgenius/dify-docs) - [x] I understand that this PR may be closed in case there was no previous discussion or issues. (This doesn't apply to typos!) - [x] I've added a test for each change that was introduced, and I tried as much as possible to make a single atomic change. - [x] I've updated the documentation accordingly. - [x] I ran `dev/reformat`(backend) and `cd web && npx lint-staged`(frontend) to appease the lint gods
yindo added the pull-request label 2026-02-21 20:50:46 -05:00
yindo closed this issue 2026-02-21 20:50:46 -05:00
Sign in to join this conversation.
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: langgenius/dify#32111