Let the agents connect into your real systems through MCP
Model Context Protocol · read & write in place
MCP is the heart of Samoe · the agents don't just fetch text to reply with, they connect into your Express, Excel, databases, and APIs and actually key, create documents, and update state. Your data stays put.
- read–write
- in place, not just fetch-and-reply
- data stays
- never leaves your systems
- DIY
- build your own MCP server
from samoe_mcp import tool, serve
@tool("create_invoice")
def create_invoice(order_id: str) -> dict:
# เขียนเข้า Express จริง ไม่ใช่ mock
return express.invoices.create(order_id)
serve(name="company-erp")
# น้องเรียก create_invoice แล้วเอกสารเกิดจริงReaching out to your tools
Connect Express, Excel, databases
Use the MCP servers we provide, or write your own to the standard spec. Register in the Console and the agents can call it immediately · no data leaves.
- Prebuilt MCP servers
- Or write your own
- Register in the Console
- Data stays put
Per-tool scopes, fully auditable
Define exactly which tools each agent may call; every call is logged and traceable in the Console · built for sensitive company data.
- Per-tool scopes
- Logged calls
- Replayable traces
- No data exfiltration
Teams already building on Samoe
Keying time removed · agents pull work from the existing system over the API and act on it.
Reached the TED Fund final round · 3 pilot customers · figures from real usage
Connect the agents to your systems via MCP
Read the MCP server guide in the docs, or reach me on LINE.
A harness that acts in real systems · not a chatbot over an API