Skip to content

Fix Zen MCP HTTP transport and Qdrant port config#2

Merged
akbad merged 2 commits intomainfrom
fix/zen-qdrant-mcps
Dec 24, 2025
Merged

Fix Zen MCP HTTP transport and Qdrant port config#2
akbad merged 2 commits intomainfrom
fix/zen-qdrant-mcps

Conversation

@akbad
Copy link
Copy Markdown
Owner

@akbad akbad commented Dec 24, 2025

Changes

tools/scripts/start-zen-http.py

  • Enable stateless mode for StreamableHTTPSessionManager: Zen's tool calls are independent so no session state is needed
  • Add timeout_graceful_shutdown=5 to uvicorn call to prevent HTTP keep-alive connections from keeping the server hanging when it's killed (e.g. via stop-beehive)

tools/scripts/set-up-tools.sh

  • Fix env var name: FASTMCP_PORTFASTMCP_SERVER_PORT (since FastMCP uses FASTMCP_SERVER_ prefix)

Rationale

Switch to stateless mode for Zen

  • Zen MCP is inherently stateless:
    • Each clink call is independent
    • No conversation memory is kept
  • Stateful mode was a choice made in hasty error when written
    • It requires session handshake before tool calls, which adds unnecessary complexity for a simple pass-through service
    • Breaks clients (like Claude Code) expecting simple request/response
  • Solution: stateless=True gives each request a fresh transport, matching Zen's nature

FASTMCP_SERVER_PORT rename

Fixes an error caused by a discrepancy between upstreams:

  • qdrant/mcp-server-qdrant README documents FASTMCP_PORT, but this is a mistake
  • FastMCP's actual ServerSettings uses env_prefix="FASTMCP_SERVER_"

@akbad akbad merged commit 9cea7a5 into main Dec 24, 2025
@akbad akbad deleted the fix/zen-qdrant-mcps branch December 24, 2025 22:12
akbad added a commit that referenced this pull request Mar 12, 2026
* Add stateless mode and 5s timeout for graceful shutdown to Zen startup script
* Fix Qdrant listening port startup flag to match FastMCP upstream of `mcp-server-qdrant`
akbad pushed a commit that referenced this pull request Apr 3, 2026
Ranks all 7 platforms: Hermes Agent (#1), OpenHands (#2), Letta (#3),
CoPaw (#4), OpenFang (#5), Memoh (#6), OpenClaw (#7). Recommends
Hermes + OpenHands as optimal integration stack. Includes head-to-head
matrix, top 10 feature merge ideas, tradeoff analysis, and phased
next steps.

https://claude.ai/code/session_01Ta99Gdof2dBsruDBdNRFZz
akbad added a commit that referenced this pull request Apr 3, 2026
* Add agent framework evaluation: Bureau vs OpenClaw, CoPaw, Hermes, Memoh, LettaBot, LocalAGI

Compares Bureau's competitive position against six always-on multi-channel
assistant frameworks. Documents two-track development plan: Track A
(multi-channel expansion) and Track B (dev tool improvements) with
concrete sprint ordering.

https://claude.ai/code/session_01Ta99Gdof2dBsruDBdNRFZz

* Add integration analysis reports for CoPaw, Letta/LettaBot, Memoh, OpenFang

Four detailed reports comparing Bureau integration fit across memory
architecture, security model, channel support, SWE features, and
autonomous capabilities. Hermes, OpenClaw, and OpenHands reports
still in progress.

https://claude.ai/code/session_01Ta99Gdof2dBsruDBdNRFZz

* Add high-impact feature merge brainstorms to CoPaw, Letta, Memoh reports

Each report now includes a Section 12 with 8-12 creative, technically
specific ideas for Bureau integration that leverage the unique
combination of both platforms.

https://claude.ai/code/session_01Ta99Gdof2dBsruDBdNRFZz

* Add high-impact feature merge brainstorms to Memoh and OpenFang reports

Memoh: 12 ideas including per-role container isolation, parallel
debugging swarms, and container-isolated MCP server vetting.
OpenFang: feature merge section with autonomous Hands + Bureau
coding role combinations.

https://claude.ai/code/session_01Ta99Gdof2dBsruDBdNRFZz

* Add OpenHands integration analysis report

Covers Docker sandbox architecture, 77.6% SWE-bench performance,
event stream state management, and integration proposal with Bureau
as orchestrator and OpenHands as sandboxed execution backend.

https://claude.ai/code/session_01Ta99Gdof2dBsruDBdNRFZz

* Add high-impact feature merge brainstorms to OpenHands report

11 ideas including verified code reviews via sandbox execution,
step-gated editing with test rollback, multi-model tournament
execution, and real attack testing in Scrimmage Mode sandboxes.

https://claude.ai/code/session_01Ta99Gdof2dBsruDBdNRFZz

* Add Hermes Agent and OpenClaw integration analysis reports

Hermes: 5-layer memory architecture, autonomous learning loop,
6 channels, 6 execution backends. Rated strongest integration
candidate for Bureau.

OpenClaw: 23+ channels, 346k stars, companion apps, voice/canvas.
Maximalist option with highest reward/risk. 9 CVEs in March 2026.

https://claude.ai/code/session_01Ta99Gdof2dBsruDBdNRFZz

* Add high-impact feature merge brainstorms to Hermes Agent report

12 ideas including role-scoped evolving memory, Assess Mode with
episodic recall, self-improving skills via feedback loop, zero-
clarification dispatch, and cross-backend execution routing.

https://claude.ai/code/session_01Ta99Gdof2dBsruDBdNRFZz

* Add high-impact feature merge brainstorms to OpenClaw report

10 ideas including voice-driven code review via Talk Mode, Live
Canvas diff theater, companion app Micro Mode approvals, ClawHub
skill transpiler, and iMessage CI/CD war room.

https://claude.ai/code/session_01Ta99Gdof2dBsruDBdNRFZz

* Add synthesis README with ranked integration recommendations

Ranks all 7 platforms: Hermes Agent (#1), OpenHands (#2), Letta (#3),
CoPaw (#4), OpenFang (#5), Memoh (#6), OpenClaw (#7). Recommends
Hermes + OpenHands as optimal integration stack. Includes head-to-head
matrix, top 10 feature merge ideas, tradeoff analysis, and phased
next steps.

https://claude.ai/code/session_01Ta99Gdof2dBsruDBdNRFZz

---------

Co-authored-by: Claude <noreply@anthropic.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant