Repository navigation
docs(integrations): add a Microsoft Agent Framework memory workflow (rehost #5959) - #6017
Conversation
[Medium impact] The PR appears safe to merge; no actionable new issue was established. SummaryAdds a Microsoft Agent Framework guide and a Python example for explicit saves and on-demand recall.
No actionable new issue was established. The live suite was not run. DiagramsequenceDiagram
participant App as Python application
participant Flow as Recall workflow
participant MCP as Local MCP server
participant Worker as Claude-Mem worker
App->>MCP: Save note with project and fingerprint
MCP->>Worker: Save memory
Worker-->>App: Saved or existing observation ID
App->>Flow: Select project, query, and ID
Flow->>MCP: Search project
MCP->>Worker: Fetch compact index
Flow->>MCP: Read scoped timeline and observation
MCP->>Worker: Fetch selected context and details
Flow-->>App: Index, timeline, and details
Reviews (2) · Last reviewed commit: "fix: preserve literal project keys and i..." · Reviewed by Greptile |
There was a problem hiding this comment.
Cursor Bugbot has reviewed your changes using high effort and found 1 potential issue.
❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.
Reviewed by Cursor Bugbot for commit d4a86d4. Configure here.
| async def find_note(request: SelectedNote, ctx: WorkflowContext[SelectedNote]) -> None: | ||
| index = text_content(await memory.call_tool( | ||
| "search", project=request.project, projects=json.dumps([request.project]), query=request.query, type="observations", limit=20, | ||
| )) |
There was a problem hiding this comment.
Search ranking breaks older-note test
Medium Severity
recall_note calls search with limit=20 and no orderBy, so the worker keeps its default relevance ranking. The live test then inserts 25 later notes and expects the original exact-title match to vanish from that page. Relevance does not page by age, and the original title is an exact query match, so the advertised verification run can fail even when ID-based timeline and get_observations still work.
Additional Locations (1)
Reviewed by Cursor Bugbot for commit d4a86d4. Configure here.


Rehost of #5959 by @Afloat16, who did all of this work. Their commits are cherry-picked unchanged with
git cherry-pick -x, so they stay the commit author on main; the original PR has maintainer edits off, so this branch carries the one-linedocs/public/docs.jsonmerge with the framework guides that landed first.Closes #5922
🤖 Generated with Claude Code
https://claude.ai/code/session_019FbY8c2BSHbdhJhnWYMGaY
Note
Low Risk
Documentation and example-only changes; no modifications to the worker, MCP server, or production runtime paths.
Overview
Adds Microsoft Agent Framework as a documented integration: a new Get Started nav entry and
microsoft-agent-framework.mdxexplain how to useMCPStdioToolwith the existing Claude-Mem MCP worker for explicit save and on-demand recall (no capture hooks or model API).The new
examples/microsoft-agent-frameworkpackage shipsmemory_workflow.py(CLIsave/recall, two-executorWorkflowBuilderflow, project-scoped MCP calls withplatformSource: microsoft-agent-frameworkand SHA-256 dedupe fingerprints), pinnedrequirements.txt, andtest_memory_workflow.py(live worker tests for identity, project isolation, worktrees, failures, and CLI diagnostics). The doc includes a disposable-worker bash block to run the suite safely.Reviewed by Cursor Bugbot for commit d4a86d4. Bugbot is set up for automated code reviews on this repo. Configure here.