mythos mcp exposes Mythos as a local Model Context Protocol stdio server.
It does not:
- start an HTTP daemon
- open a port
- call Anthropic, OpenAI, or DeepSeek
- duplicate SWD logic
The MCP client launches mythos mcp as a subprocess and calls tools over stdin/stdout.
mythos mcp config generic
mythos mcp config claude
mythos mcp config cursor
mythos mcp config cursor --jsonThe generic shape is:
{
"mcpServers": {
"mythos-router": {
"command": "mythos",
"args": ["mcp"]
}
}
}Run the MCP client from the repository you want Mythos to guard, or use a project-scoped MCP config when the client supports it.
| Tool | Purpose |
|---|---|
swd_dry_run |
Preview external-agent file actions without writing files or receipts |
swd_apply |
Apply file actions through SWD, verify disk state, roll back failed writes, and write receipts |
receipts_list |
List recent local SWD receipts |
receipts_show |
Show a receipt as JSON or PR-ready Markdown |
receipts_verify |
Re-check current files and receipt integrity |
skills_list |
List visible project/global skill packs |
skills_check |
Validate skill packs |
Most users should let their MCP client call the server. For a low-level smoke test:
printf '%s\n' \
'{"jsonrpc":"2.0","id":1,"method":"initialize","params":{"protocolVersion":"2025-06-18","capabilities":{},"clientInfo":{"name":"manual-smoke","version":"0.0.0"}}}' \
'{"jsonrpc":"2.0","id":2,"method":"tools/list"}' \
| mythos mcpYou should see swd_dry_run, swd_apply, receipt tools, and skill tools in the response.