Protocol

Open MCP endpoint — connect any client, or none.

Vectis exposes your tenant’s data and write-back surface over the open-standard Model Context Protocol. Point any compliant client at your instance — Vectis itself never calls an AI model. Compatible with Claude, Cursor, ChatGPT, Continue, and any self-hosted client.

Read the setup guide Start a Vectis workspace

Per-tenant isolation

Tokens are sha256-hashed and stored only in your tenant’s database. A token issued at acme.mspvectis.com cannot authenticate at any other tenant — the hash literally does not exist there.

Bring your own client

Any MCP-compliant client works: Claude Desktop, Cursor, ChatGPT custom GPT Actions, Continue with a local Ollama model, LibreChat, open-webui. Same endpoint, same token.

No vendor lock-in

Vectis ships no AI model and calls no AI vendor. If your policy forbids third-party AI processors, point a self-hosted client at the same endpoint and your data stays on your machine.

Supported clients

The endpoint speaks Model Context Protocol (Streamable HTTP transport). Each client below has documented setup instructions in the setup guide.

Claude Desktop
via mcp-remote bridge
Cursor
native HTTP MCP
Continue (Ollama / local)
stays on your machine
Claude API connector
static-bearer integration
ChatGPT custom GPT
OpenAPI shim provided
LibreChat / open-webui
MCP-compliant clients

Sample prompt → tool call → result

Prompt

“Which customers blew up overnight?”

Client → Vectis

tools/call list_alerts {
  "severity": "critical",
  "status": "open"
}

Vectis → Client

{ "alerts": [
  { "title": "RAID degraded",
    "customerName": "Acme Inc.",
    "severity": "critical" }
] }

Every tool call lands in the tenant audit log with the calling user, redacted arguments, and the token prefix. Operators can revoke tokens instantly from /account/mcp.