Scipio ERP 4.0 ships no agent of its own. It gives the one you already use the hands: every application exposes its services as MCP tools on the server’s own endpoint, and the agent you connect operates the system the way a user with that login would. Claude Code, Claude Desktop, Cursor and VS Code connect with one command.
Bring your agent
Connect the agent you already use to Scipio ERP over MCP. Every application ships an MCP server; a token is a user login; every call is on record.
One command.
Then ask.
- Create a token. Admin › Agent Access › Tokens. A token is a user login: the agent gets that user's permissions, and not one more.
- Add the server to your agent. Every application has its own endpoint under its webapp; the admin hub on the right reaches all of them.
- Ask in plain words. The agent finds the tools, calls them with that login, and every call writes one audit row.
claude mcp add --transport http scipio \ "https://<host>/admin/mcp" \ --header "Authorization: Bearer <token>"
Fifteen servers. 98 tools.
Each application's server exposes that application's services as tools, with the input and output schema read from the service definition. Pick an application to see its tools.
-
Shop
shop9 tools -
Order
order7 tools -
Catalog
catalog3 tools -
CMS
cms17 tools -
Content
content4 tools -
Accounting
accounting7 tools -
Warehouse
facility5 tools -
Manufacturing
manufacturing5 tools -
CRM
marketing8 tools -
Party
party4 tools -
HR
humanres6 tools -
Work Effort
workeffort5 tools -
Solr
solr3 tools -
Setup
setup3 tools -
Admin
admin2 tools
A token is a login.
Not a master key.
Your agent signs in as a Scipio user and gets that user’s permissions, not one more.
Default deny. One audit row per call. A separate permission for code. Revoke a token and it is gone at once.
- Default denyevery gate must allow a call
- 1 row per callallowed or denied
- 90 daystoken expiry, 365 at most
- Same rightsas the user behind the token
Agent Access · Audit
1,284 calls today · 1 denied| Time | Token | Tool | ms | Result |
|---|---|---|---|---|
| 09:12:04 | scp_b7e2 | cms_page_create | 188 | OK |
| 09:12:41 | scp_b7e2 | cms_page_add_version | 97 | OK |
| 09:13:02 | scp_b7e2 | cms_page_publish | 74 | OK |
| 09:31:04 | scp_7f21 | order_find | 212 | OK |
| 09:31:06 | scp_7f21 | order_change_status | 148 | OK |
| 09:31:06 | scp_7f21 | order_change_status | 131 | OK |
| 09:31:07 | scp_7f21 | order_change_status | 140 | OK |
| 09:31:09 | scp_8a02 | cms_script_update | 4 | DENIED · MCP_CODE_WRITE |
| 09:52:12 | scp_91c3 | shop_checkout | 330 | OK |
| 10:04:55 | scp_c93d | bom_get | 61 | OK |
Every call writes one row, allowed or denied, with sensitive arguments redacted. HTTPS only · rate limits per token · self-hosted.
The screens behind it.
Tokens, the audit log, the servers and the skills live in Admin › Agent Access. This is the real screen, in the Aurora theme.

Read on.
Quickstart
Create a token, connect Claude Code, Claude Desktop, Cursor or VS Code, make the first call.
Security
The token model, the permissions, the policy, the audit log, expiry and caps.
Extending
Add a tool, a service tool, a server profile, a provider or a skill of your own.
MCP annotations
@McpServer, @McpServiceTool, @McpTool and the rest, attribute by attribute.
See it run on your own data.
A walk through the application with the people who build it, on your catalogue and your processes. Or bring your agent and try the tools yourself.