Server Management

Register a Server

Via the dashboard

Navigate to MCP Gateway → MCP Servers and click Register Server. Provide:

Name — Human-readable identifier (e.g. "GitHub MCP")
Server URL — The MCP server's SSE endpoint
Connector type — Custom, Salesforce, Copilot, ServiceNow, Cursor, Windsurf, or Claude Code
Description — Optional context for your team

New servers start in the PENDING_APPROVAL state. Activate them via the dashboard or API after review.

Via the API

curl -X POST /api/v1/mcp/servers \
  -H "Authorization: Bearer {api_key}" \
  -d '{
    "name": "GitHub MCP Server",
    "server_url": "https://mcp.github.com/sse",
    "connector_type": "custom"
  }'

The response contains the server id; the governed endpoint agents connect to is POST /api/v1/mcp/proxy/{id}.

For any of the 11 connector templates (Salesforce, Microsoft Copilot, ServiceNow, Slack AI, Teams Copilot, Zendesk AI, Intercom Fin, AWS Bedrock, Azure AI, Cursor, Claude Code), use the template endpoint to automatically create the server with recommended policies:

curl -X POST /api/v1/mcp/servers/from-template \
  -H "Authorization: Bearer {api_key}" \
  -d '{ "template": "salesforce" }'

Tool Discovery Scan

After registration, scan the server to discover its tools:

POST /api/v1/mcp/servers/{id}/scan

The scan connects to the MCP server, calls tools/list, and for each tool:

Registers the tool with name, description, and input schema
Hashes the tool description (SHA-256) for rug-pull detection
Automatically classifies risk based on tool name and description:

Risk Level	Triggers
CRITICAL	`exec`, `shell`, `command`, `eval`, "execute arbitrary"
HIGH	`write`, `delete`, `remove`, `drop`, `modify`, `update`, `insert`
MEDIUM	Default for unrecognized tools
LOW	`read`, `get`, `list`, `search`, `query`, "read-only"

Poisoning detection — if a tool's description hash has changed since the last scan, it is marked is_poisoning_suspect: true

Server Status

Status	Meaning
`PENDING_APPROVAL`	Newly registered, not yet active
`ACTIVE`	Operational — tool calls are proxied
`BLOCKED`	Manually or emergency-stopped — all tool calls rejected
`SCANNING`	Tool discovery in progress
`ERROR`	Server unreachable or scan failed

Monitoring

The Command Center displays MCP KPIs in near real time:

Tool calls and block rate
Active vs. total servers
Number of pending approvals
Poisoning alerts
Server errors

MCP tool calls appear in the Live Activity Feed as MCP_TOOL_CALL trace type.