Describe it. Your AI builds it.
It runs forever.
MCPWorks is an open-source AI agent runtime. Tell your AI assistant what to automate. It builds a persistent agent — with scheduling, webhooks, encrypted state, and communication channels — that runs 24/7 using 70-98% fewer tokens than standard MCP.
Self-host free forever under BSL 1.1. Or join the Cloud waitlist. Same software either way.
$ claude "build me an ETH price agent"
✓ Created agent ethwatch
✓ Created function check-price
→ Python, monitors ETH price feed
✓ Scheduled every 5 minutes
✓ Connected to #alerts on Discord
✓ AI engine: Kimi K2.5 (your key)
✓ Agent running. First check in 5m.
3 hours later, on your phone:
ethwatch: ETH dropped 6.8% in the last
hour. Correlated with SEC filing on
staking regulations. Want me to dig
deeper?Works with any client that supports the Model Context Protocol (MCP) — the open standard adopted by Google, Microsoft, and governed by the Linux Foundation
What can you build with MCPWorks?
MCPWorks agents are persistent, autonomous processes that run 24/7. Each agent has its own container, schedule, encrypted state, and communication channels. Describe the automation to your AI — it builds the agent for you.
Automation · Cron · Discord
Competitor Price Monitor
Track competitor pricing every hour. Get a Discord alert when anything changes. No AI needed — pure automation, $0 tokens.
"Build me an agent that monitors competitor pricing and alerts me on Discord."
Autonomous · Webhook · Discord
GitHub Issue Triage
When a new issue lands, the agent reads it, labels it by severity, drafts a response, and posts a summary to Discord. Uses your own OpenAI key.
"Build an agent that triages GitHub issues and labels them automatically."
Automation · Cron · Email
Daily Revenue Report
Pull yesterday's Stripe revenue, format a clean summary, email it to your team at 7am. Runs every morning. No AI needed.
"Build me an agent that emails daily Stripe revenue summaries at 7am."
How do you build an AI agent with MCP?
Building an AI agent with MCPWorks takes three steps: clone the repo, describe what you want, and your agent runs autonomously. Your AI assistant handles the code — you describe the outcome.
Clone or connect
Self-host with docker compose up or connect to MCPWorks Cloud. Add the MCP server to your AI assistant's config — one JSON block.
Describe what you want
Tell your AI assistant what to automate in plain language. It creates the agent, writes the functions in Python or TypeScript, sets the schedule, and connects your channels.
Your agent runs 24/7
The agent runs on a schedule, responds to webhooks, and messages you on Discord, Slack, or email. Update it anytime by talking to your AI.
Why do MCPWorks agents cost less to run?
Standard MCP tool calls load schemas into context, burning tokens on boilerplate the model already understands. MCPWorks agents write code in a secure sandbox instead — data stays in the sandbox, never enters context. Same capabilities, 70-98% fewer tokens.
40,000+
tokens per tool call with standard MCP
70-98%
reduction with MCPWorks code-mode
$0
to self-host. Your hardware, your savings.
Free to self-host. Cloud from $179/mo. See pricing →
Frequently Asked Questions
What is MCPWorks? +
MCPWorks is an open-source AI agent runtime built on the Model Context Protocol (MCP). The full platform — agent runtime, nsjail code sandbox, scheduling, webhooks, encrypted state, and communication channels — is released under BSL 1.1 (converts to Apache 2.0 after 4 years). Self-host for free with no limits, or use MCPWorks Cloud for managed hosting.
How is MCPWorks different from LangChain or CrewAI? +
MCPWorks is infrastructure, not a framework. LangChain and CrewAI provide SDKs — you still need to host, schedule, persist state, and manage secrets yourself. MCPWorks provides the runtime: your AI builds the agent through MCP tools, and MCPWorks handles the container, cron, webhooks, encrypted state, and communication channels.
Can agents run without an LLM? +
Yes. Automation mode runs functions on cron schedules and responds to webhooks with zero token cost. Autonomous mode adds an LLM brain for reasoning. BYOAI — connect your own API key from 14+ providers. MCPWorks never hosts or proxies AI.
Your first agent is one conversation away.
Clone the repo and self-host in 5 minutes. Or join the Cloud waitlist for first access when managed hosting launches.
Get Cloud launch updates and token efficiency research:
Last updated: