Claude Code v2.1.259 adds managedMcpServers and headless permission-prompts none
cliClaude Code 2.1.259 introduces a `managedMcpServers` managed setting so organizations can push HTTP/SSE MCP servers to every user (command-based entries are skipped), and a new `--permission-prompts none` flag for unattended headless hosts that auto-denies anything that would prompt while the active permission mode still applies. Adds `--json` output to `claude plugin validate`, GitLab `glab` MR recognition, and many fixes around concurrent sessions, MCP startup errors, workt
An agent operator can now centrally push HTTP/SSE MCP servers to every Claude Code user via `managedMcpServers` and run fully unattended headless sessions with `--permission-prompts none` that auto-deny anything requiring a prompt.
Meaningful capability delta for agent orchestrators: enterprise-managed MCP distribution + true unattended headless mode, plus real MCP-startup error surfacing.
npm i -g @anthropic-ai/claude-code@2.1.259# Unattended headless run: auto-deny anything that would prompt
claude --permission-prompts none -p "Refactor the auth module and run tests"
# Machine-readable plugin validation
claude plugin validate ./my-plugin --json
# Managed MCP servers example (managed-mcp.json pushed by org policy)
cat <<'JSON'
{
"managedMcpServers": {
"corp-docs": { "type": "http", "url": "https://mcp.corp.example/docs" }
}
}
JSON