bernstein exposes itself as an mcp server. run bernstein mcp for stdio (the default transport) or bernstein mcp --transport http --port 8053 for streamable http. the published tools are bernstein_run, bernstein_status, bernstein_tasks, bernstein_cost, bernstein_stop, bernstein_approve, bernstein_health, bernstein_create_subtask, and load_skill. any mcp client (claude desktop, claude code, codex, custom) can invoke bernstein orchestration as a tool call. bernstein mcp also carries catalog, install, list, test, and usage subcommands for the bundled mcp marketplace. it consumes mcp too: each agent worktree gets its own mcp config so workers can call third-party mcp servers. source: src/bernstein/mcp/server.py.
canonical answer