Launching today

mcp-writecheck
Test your MCP server's writes for lost updates and retries
1 follower
Test your MCP server's writes for lost updates and retries
1 follower
Agents write to real systems through MCP servers. mcp-writecheck tests those write tools: two writers on one record, a retry after a lost answer, a batch that fails part-way. One result per requirement of MWS 0.9, an open draft. CLI, GitHub Action, Apache-2.0.





Free
Launch Team

AWS Activate CreditsCredits that fuel your startup’s next milestone
Promoted
Hey Product Hunt, Joshua here, founder of IsoKron.
We built this because we kept running into write-safety issues while using a lot of agentic workflows. An agent writes to a real system through an MCP server, gets "success" back, and the change is gone. Or a call times out, the agent retries, and the write happens twice. I spent ten years in support engineering at Ring, and the worst problems were always the ones where every system said "done" and the data said otherwise. Same thing, just with agents now.
So we tested it. Two agents editing one record through GitHub's MCP server lost one of the edits in 19 of 30 tries, and both were told it worked. Then we read 100 popular MCP servers: only 7 can say "I don't know if that write happened" in a way a program can read.
mcp-writecheck tests your own server for this in a test environment, as a CLI or a GitHub Action. It's a preview. It checks against MWS 0.9, an open draft of 37 requirements, open for comment until Dec 16.
Try it on our sample server in one command:
npx -p @mcp-write-safety/writecheck -p @mcp-write-safety/tiny-tracker mcp-writecheck demo
Full disclosure: IsoKron also does MCP audits. The tools, spec and data are free.
What does your MCP server return when it can't tell if a write landed?