io.github.cloudlabworks/mcp-failure-lab
MCP Failure Lab
Is the MCP Failure Lab MCP server down? No. It answered initialize and tools/list in the probe of 2026-11-03.
Deliberately faulty MCP server for testing how agent clients handle tool errors and disconnects.
alive probed 2026-11-15 · unchanged since 2026-11-03
- Run it
- docker run -i --rm ghcr.io/cloudlabworks/mcp-failure-lab:0.1.2 (stdio, probed in a container with the network off)
- Host
- ghcr.io
- Reports itself as
- mcp-failure-lab 0.1.0
- Capabilities
- tools
- Instructions
- none
- Of a 1M window
- <0.1 % of a 1M context window
- Count
- 125 tokens in Claude Opus 5.5, exact, measured 2026-11-03. With the Claude Code prefix mcp__…__ on each tool name: 139. The o200k_base estimate is 65. Counted again when the tool definitions change. Method.
- Repository
- https://github.com/cloudlabworks/mcp-failure-lab
- Website
- https://github.com/cloudlabworks/mcp-failure-lab
- Registry
- active, updated 2026-11-03 · registry entry
Most expensive tools
The Claude column scales the o200k estimate of each tool by this server's ratio of exact to estimate, 1.92×.
| Tool | ≈ Claude tokens | o200k estimate | Share of this server |
|---|---|---|---|
| lab_probe | 125 | 65 | 100 % |
Compare this server with others.
More servers
Other registry entries on ghcr.io, the ones next to this name; all of them.
- CityEngine Cloud io.github.cgajs/cityengine-mcp · alive
- WPGuard MCP io.github.cgallic/wpguard-mcp · not probed
- tollbooth io.github.chrzanowy/agent-tollbooth · not probed
- io.github.claytono/go-unifi-mcp · crashed on start
- codeDB Lens io.github.codedb-io/codedb-lens · needs args
- io.github.coe0718/deskbrid · alive
- CMDOP Machine Care io.github.commandoperator/cmdop-care · alive
- io.github.constantinemakoid/celmis · not probed
Servers that mention failure.
- AFMR Discovery ai.afmr/discovery · alive
- ai.imaginode/imaginode · needs auth
- ai.nullary/nullary · alive
- ai.plith/plith · alive
- Agent Rescue Desk Production Triage app.netlify.agent-rescue-desk/production-triage · alive
- BuyerAware com.buyerawareapp/buyeraware · alive
- com.clauxel.a2areplayreceipt/a2areplayreceipt-mcp · no DNS
- Claidex MCP com.claidex/failure-intelligence · not found
Badge for the README
The result of the latest probe, the tool count and the token cost. The feed has an entry for every probe that changed them.
A mail when this server changes
Stops answering, comes back, tool set or cost changes. One confirmation mail first, a stop link in every mail.
For the publisher
- The server answers initialize and tools/list. Nothing to fix.