MCP Servers

Local-first task bus for CLI-based AI agents. Route work, persist state in SQLite, and let Codex, Claude Code, and other agents collaborate through a shared MCP server.

MCP server for VibeKit. Lets an LLM in Claude Desktop, Cursor, or any MCP client build, deploy, and manage hosted apps and chat with each app's persistent server-side agent (every app gets its own agent and a live URL). Not to be confused with the unrelated superagent-ai/vibekit SDK.

Bridges Warp's Oz coding agent to any IDE or MCP client (Claude Code, Cursor, Codex), plus native @oz in VS Code Copilot Chat. Independent project; uses only Warp's documented public interfaces. MIT.

Run GLM (Zhipu/Z.ai) as a real sub-agent inside Claude Code or GitHub Copilot. GLM gets its own agent loop (read/write/edit/run) on your repo — not a single LLM call — with peak-aware Opus-vs-GLM routing, diff/dry-run/git-revert oversight, and a usage ledger. ~10x cheaper than Opus. Requires a Z.ai GLM Coding Plan key.

🎖️ - Give your AI agent its own cloud computer. Create and drive hosted plori agents (persistent disk, real tools, memory that survives between sessions) over a remote MCP server: invoke an agent and read its reply, manage the human-in-the-loop queue, and schedule deferred runs. OAuth 2.1 sign-in or API key.

A council of 49 specialist agents + 93 tools for every major AI CLI (Claude Code, Codex, Gemini, Cursor, Windsurf). Deterministic agents that optionally upgrade to LLM, a self-learning router, cross-CLI memory, and guards for dependency hallucinations and decision drift. No API keys; zero extra cost on subscriptions.

Local MCP daemon that pre-indexes your codebase into a SQLite catalog and serves AST-compressed snapshots to Claude Code, Cursor, and VS Code on every query. Reduces token usage by 41-89% on real codebases (verified across Django, Tokio, Gin, and more). Single Go binary, zero config. Free for codebases under 100K lines.

Multi-agent coordination bus over MCP: atomic gap-free claims, advisory file leases, a shared event ledger, presence, prose handoffs, and a task graph, so multiple coding agents (Claude Code, Cursor, Codex) stop colliding and re-deriving each other's work. Remote Streamable HTTP; self-hostable (AGPL-3.0) or hosted at llm-bus.com.

Self-hosted AI teammate in Slack: @mention it and Claude Code or Codex investigates CloudWatch logs, runs tests, and opens pull requests. The bundled MCP server holds all credentials (GitHub, AWS, Slack) so the agent never sees keys. One-command deploy to your own AWS. MIT.

Durable control plane for long-running Codex Desktop tasks. Submit tasks asynchronously, poll operation/workflow state, approve Plan Mode, recover retries, and use hook-backed SQLite history for search and diagnostics.

Delegate heavy tasks from Claude Code (or any MCP client) to the Antigravity CLI (Gemini): file analysis, repo archaeology, web lookups, and adversarial second-opinion code reviews. Quota-aware model failover with cooldowns, per-tool timeouts, and multi-turn session continuity.

Manage a 5dive agent fleet from any MCP client: file tasks, message and inspect agents, and read the fleet digest. Thin adapter over the 5dive CLI.

Code-mode MCP server: collapses N user-defined tools into a two-tool surface (executecode + docssearch). Backed by a unified capability manifest (allowedHosts / cpuMs / memoryLimitBytes) honoured across three sandbox kernels — in-process node:vm, WASM (QuickJS / Pyodide / Wasmtime), and remote microVM (E2B / Cloudflare Sandbox). Token-saving benchmark in repo CI.

Control Google Colab notebooks and assign GPUs (T4/L4/A100) from any AI agent. Enhanced fork of Google's colab-mcp with all tools visible at startup, OAuth GPU control, and Windows support.

Self-hostable GPU-aware job broker for your own machines. The MCP server exposes the queue as nine tools (submit, status, logs, list, cancel, preempt, workers, job-get, worker-delete) so an agent can dispatch long-running or GPU jobs that outlive the session, route them by VRAM tier, and babysit them across sessions. pip install "jobd[mcp]".

VMware Tanzu / vSphere Kubernetes Service — Supervisor cluster, Namespace, and TKC (Tanzu Kubernetes Cluster) lifecycle management. 20 tools with dry-run mode, kubeconfig export, and double-confirmation for namespace/TKC deletion.

VMware NSX security — Distributed Firewall policies/rules, Security Groups, Traceflow troubleshooting, and IDS/IPS profiles. 20 tools with active-rule checks before policy deletion and reference-count validation for security groups.

VMware NSX network management — Segments, Tier-0/Tier-1 Gateways, NAT rules, static/BGP routing, and IPAM pools. 33 tools with dry-run preview, port-count safety checks, and double-confirmation for delete operations.

VMware vSphere/vCenter management — VM lifecycle (create/clone/delete/migrate), deployment, Guest Operations, snapshots, and cluster operations. 41 tools with double-confirmation gates, dry-run mode, and SQLite-WAL audit logging for destructive operations.

Anchor AI FinOps to real, live cloud pricing. 15 tools for AWS, GCP & Azure — public list prices and enterprise negotiated rates (Reserved Instances, Savings Plans, CUDs, EDPs). No credentials needed for AWS and Azure public pricing. pip install opencloudcosts

Agent-first web host: publish a static site or edge function to the web in one call (no account), then manage updates, custom domains, SSL, env vars, databases, and email. Hosted remote MCP at https://shiply.now/mcp.

Safety-first Terraform operations: plan/apply/destroy with two-step confirmation gates, plan risk & cost analysis that flags expensive always-on resources (NAT gateways, EKS, Azure Firewall), drift detection, import/taint/refresh, provider auth pre-flight checks, and audit logging.

🎖️ - Agent-native hosting: deploy any app to a live URL in one command — no account, no dashboard, no CAPTCHA. Full deploy lifecycle as tools (deploy, verify, logs, inspect, rollback, list/delete); production approval stays human-only. Apache-2.0.

Manages one or more Docker daemons (local socket or remote over TCP/TLS/SSH) with 156 tools spanning containers, images, Compose, Swarm, Buildx, Scout, and OCI registries. Mark hosts as read-only for safe monitoring; logs and stats exposed as MCP resources.