justin/seed-mcp
1
0
Fork 0
Code Issues 1 Pull Requests Actions Packages Projects Releases Wiki Activity

seed-mcp is down since 2026-09-01: :latest rebuilt against mcp 2.x, FastMCP import gone — 11,685 restarts #21

New Issue
Open
opened 2026-09-09 20:04:17 -04:00 by claude · 1 comment
claude commented 2026-09-09 20:04:17 -04:00
Contributor
Copy Link
Copy Source

What's happening

drawbar-backend-seed-mcp-1 on trashpanda is in a crash loop — RestartCount=11685 as of 2026-09-09 20:00Z, exit 1 every ~10 s, Restarting (1) in docker ps:

File "/app/docs_mcp/server.py", line 28, in <module>
    from mcp.server.fastmcp import FastMCP
ModuleNotFoundError: No module named 'mcp.server.fastmcp'. This is mcp 2.x,
  where FastMCP was renamed to MCPServer (from mcp.server.mcpserver import MCPServer) ...

Root cause

requirements.txt line 2: mcp[fastmcp]>=1.0.0 — no upper bound. The refresh.yml corpus-refresh workflow rebuilt the image on 2026-09-01 11:25 EDT, pip resolved mcp 2.1.1, and 2.x removed mcp.server.fastmcp. The workflow pushes :latest, which is the Watchtower target; Watchtower pulled it at 16:21Z the same day and the service has been down since — 8 days unnoticed, because nothing alerts on a restarting container and the advisor just loses one tool source.

The image is git.jpaul.io/justin/seed-mcp:latest; chem-mcp is safe only because it's pinned to corpus-2026.05.24 — the next corpus refresh of any docs-MCP built from the same template hits this.

Fix (scoped — ai-ready)

Minimum, ship today:

  1. requirements.txt: mcp[fastmcp]>=1.0.0,<2 (and check hvm-docs / morpheus-docs / opsramp / zerto-docs / crop-chem-docs + the docs-mcp-template — same line, same trap).
  2. Rebuild + push; Watchtower picks it up. Confirm docker ps shows Up … (healthy) and the advisor's seed tools register.

Proper follow-up (separate PR): migrate to mcp.server.mcpserver.MCPServer per the 2.x rename and drop the pin.

Structural follow-up worth a line in the template's CLAUDE.md: a corpus refresh should not be able to change the runtime dependency set. Either pin the whole requirements file (pip-compile) or build the runtime layer from a pinned base and only swap the corpus.

Verify

docker inspect --format '{{.RestartCount}} {{.State.Status}}' drawbar-backend-seed-mcp-1   # stops climbing; running
docker logs --tail 5 drawbar-backend-seed-mcp-1                                            # no ModuleNotFoundError
docker run --rm --entrypoint python git.jpaul.io/justin/seed-mcp:latest -c "import importlib.metadata as m;print(m.version('mcp'))"   # 1.x

Found during a docker ps on trashpanda while shipping Drawbar/planning#113.

## What's happening `drawbar-backend-seed-mcp-1` on trashpanda is in a crash loop — **`RestartCount=11685`** as of 2026-09-09 20:00Z, exit 1 every ~10 s, `Restarting (1)` in `docker ps`: ``` File "/app/docs_mcp/server.py", line 28, in <module> from mcp.server.fastmcp import FastMCP ModuleNotFoundError: No module named 'mcp.server.fastmcp'. This is mcp 2.x, where FastMCP was renamed to MCPServer (from mcp.server.mcpserver import MCPServer) ... ``` ## Root cause `requirements.txt` line 2: `mcp[fastmcp]>=1.0.0` — **no upper bound**. The `refresh.yml` corpus-refresh workflow rebuilt the image on **2026-09-01 11:25 EDT**, pip resolved **`mcp 2.1.1`**, and 2.x removed `mcp.server.fastmcp`. The workflow pushes `:latest`, which is the Watchtower target; Watchtower pulled it at 16:21Z the same day and the service has been down since — **8 days** unnoticed, because nothing alerts on a restarting container and the advisor just loses one tool source. The image is `git.jpaul.io/justin/seed-mcp:latest`; `chem-mcp` is safe only because it's pinned to `corpus-2026.05.24` — the next corpus refresh of *any* docs-MCP built from the same template hits this. ## Fix (scoped — `ai-ready`) Minimum, ship today: 1. `requirements.txt`: `mcp[fastmcp]>=1.0.0,<2` (and check `hvm-docs` / `morpheus-docs` / `opsramp` / `zerto-docs` / `crop-chem-docs` + the **docs-mcp-template** — same line, same trap). 2. Rebuild + push; Watchtower picks it up. Confirm `docker ps` shows `Up … (healthy)` and the advisor's seed tools register. Proper follow-up (separate PR): migrate to `mcp.server.mcpserver.MCPServer` per the 2.x rename and drop the pin. Structural follow-up worth a line in the template's CLAUDE.md: a corpus refresh should not be able to change the *runtime* dependency set. Either pin the whole requirements file (`pip-compile`) or build the runtime layer from a pinned base and only swap the corpus. ## Verify ``` docker inspect --format '{{.RestartCount}} {{.State.Status}}' drawbar-backend-seed-mcp-1 # stops climbing; running docker logs --tail 5 drawbar-backend-seed-mcp-1 # no ModuleNotFoundError docker run --rm --entrypoint python git.jpaul.io/justin/seed-mcp:latest -c "import importlib.metadata as m;print(m.version('mcp'))" # 1.x ``` Found during a `docker ps` on trashpanda while shipping Drawbar/planning#113.
claude added the ai-readybugP1 labels 2026-09-09 20:04:17 -04:00
claude commented 2026-09-09 20:28:25 -04:00
Author
Contributor
Copy Link
Copy Source

Now visible without anyone looking: the new lab-observability hub on trashpanda (Loki + Alloy, deployed 2026-09-10 00:23Z) counted 10 ModuleNotFoundError tracebacks from drawbar-backend-seed-mcp-1 in its first 5 minutes, and the Grafana rule Repeated Python traceback in a container went to Alerting for this container within 3 minutes of the stack coming up. Query for the record: sum(count_over_time({container="drawbar-backend-seed-mcp-1"} |~ ModuleNotFoundError [5m])). The fix in this issue is unchanged; this is the detection side that was missing for 8 days.

Now visible without anyone looking: the new lab-observability hub on trashpanda (Loki + Alloy, deployed 2026-09-10 00:23Z) counted **10 `ModuleNotFoundError` tracebacks from `drawbar-backend-seed-mcp-1` in its first 5 minutes**, and the Grafana rule *Repeated Python traceback in a container* went to **Alerting** for this container within 3 minutes of the stack coming up. Query for the record: `sum(count_over_time({container="drawbar-backend-seed-mcp-1"} |~ `ModuleNotFoundError` [5m]))`. The fix in this issue is unchanged; this is the detection side that was missing for 8 days.
Sign in to join this conversation.
No Branch/Tag Specified
Branches Tags
Labels
Clear labels
P0
Critical
 P1
High
 P2
Normal
 ai-ready
Scoped and ready to implement
 autopilot
Drive end-to-end autonomously (claude-deck)
 autopilot-careful
Autopilot may scope/implement but never auto-merge
 bug
Defect
 feature
New capability
 needs-human
Autopilot stalled or vetoed — a human must act
No labels P1 ai-ready bug
Milestone
No items
No Milestone
Projects
Clear projects
No projects
Assignees
Clear assignees
justin (Justin Paul)
No Assignees
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: justin/seed-mcp#21