Claude banned your account: what to do with projects and tests and where to move
Checkcraft is a desktop app for checking APIs and logs without code, with a free edition. Download, no sign-up →
Since autumn 2026 developers from Russia report Claude bans en masse. The code in the repository stays, but the checks the agent ran right in the chat disappear with the account. Here is what to keep and how not to depend on one agent.
What disappears and what stays
Everything the agent wrote into the repository stays: code, tests, instruction files. What goes away is the account, the chat history and whatever the agent checked right in the chat: curl requests, one-off scripts, lines like "looked at the response, all fine". Those checks are saved nowhere and cannot be repeated after a ban. Yet they are usually what proved a change worked.
Migration checklist
Go through the list before a ban or right after it. The key step is turning one-off chat checks into saved ones, with expected values taken from the requirements rather than from whatever the server returned.
Before the ban or right after it:
[ ] Tests the agent wrote into the repository are still there: they are your code.
[ ] Checks the agent ran right in the chat (curl, one-off scripts,
"looked at the response, all fine") are saved nowhere. List them and save
them as checks with expected values.
[ ] Move the project instruction file (CLAUDE.md) to the new client's format —
many read AGENTS.md.
[ ] Move your MCP server settings into the new client's config.
[ ] Passwords and tokens belong in environment variables, not in chat history.Where to move
MCP-capable clients such as Qwen Code and OpenCode connect external tools and work with different models, including Qwen and DeepSeek. You can switch models in an evening. Knowledge of what was checked and how is harder: if it lived only in the chat, you will have to rebuild it.
Checks separate from the agent
It is safer to keep checks where they cannot be taken away with an account or quietly rewritten to match a response. In Checkcraft checks live as files on your computer and run from saved rules: the same data gives the same result. Over MCP the agent only creates and runs them and reads the report. If the agent is banned, the checks stay, and a new agent connects to them with the same command. The connection is tested with Claude Code and Cursor; other MCP clients use the same server command. You need the Windows build 1.0.185 or later; the free version works.
{ "mcpServers": { "checkcraft": { "command": "C:\\Checkcraft\\checkcraft-mcp.exe" } } }Repeat this on your own API
Checkcraft is a desktop app that saves API response checks without code and shows what did not match. Start with a sample, then use your own sanitized data. Free runs on Windows and Linux with no time limit.
Download FreeDownload and first runNo sign-up or email needed, the archive is ready in a couple of minutes. See the edition limits.
References
vc.ru: Claude bans for developers from Russia · Qwen Code: MCP servers · OpenCode: MCP servers · Model Context Protocol
Start with the linked example and compare its result with your own test requirement.