llm-cli-gateway
A local MCP server giving one MCP client session-aware control over several AI coding-agent CLIs, while keeping each tool's native sessions and local credentials.
llm-cli-gateway lets supported MCP clients operate Claude Code, Codex, Gemini and Antigravity, Grok Build, Mistral Vibe, Cognition Devin, Cursor Agent, and configured HTTP API providers through one user-owned gateway, while preserving each tool's native CLI sessions, local credentials, durable async jobs, validation receipts and review workflows, rather than turning them into a generic chat proxy.
It persists long-running jobs with restart-safe result collection, deduplication, cancellation and sync-to-async deferral, and tracks sessions, native CLI resume paths, structured response metadata and cache telemetry, including explicit Claude cache-control support. It can run provider requests inside gateway-managed, isolated git worktrees for multi-agent review and implementation loops, backed by either a local file store or a shared PostgreSQL session manager.
It publishes CI and security-workflow status, an OpenSSF Scorecard, OpenSSF Best Practices passing status and Sigstore-signed releases, installs globally with npm or directly through npx, and is MIT licensed.