Documentation

Sorvek Bridge · MCP connection guide

Sorvek Bridge is our MCP (Model Context Protocol) server: it plugs your AI client (Claude Code, Cursor, Copilot Studio…) into your SaaS Discovery data, read-only. Your LLM can query your shadow IT inventory, your findings and your posture. It can never take an action or change anything.

Requirements

  1. 1Sorvek SaaS Discovery active for your organization.
  2. 2A Sorvek Bridge API key (type "MCP server"), generated by an admin in Settings → API keys (prefix skv_mcp_…). The key is shown only once, keep it in a secrets manager, never in a committed file.

Connect in Claude, step by step

The easiest way, no terminal: add Sorvek as a custom connector and authorize read-only access. Click through the full flow.

Other clients (terminal & config)

For Claude Code, Cursor or Copilot Studio. Server URL: https://app.sorvek.com/mcp

Claude Code

terminal
claude mcp add --transport http sorvek https://app.sorvek.com/mcp \
  --header "Authorization: Bearer $SORVEK_MCP_KEY"

Cursor

.cursor/mcp.json
{
  "mcpServers": {
    "sorvek": {
      "url": "https://app.sorvek.com/mcp",
      "headers": { "Authorization": "Bearer ${env:SORVEK_MCP_KEY}" }
    }
  }
}

If the ${env:…} interpolation fails (a known bug in some Cursor versions for remote servers), paste the key in clear text in a local, uncommitted mcp.json.

Microsoft Copilot Studio

Add an existing MCP server: transport Streamable HTTP, the URL above, authentication API key sent in the Authorization header with the value Bearer <your key>.

Claude Desktop

Easiest: use the custom connector method above (Connect in Claude), it works on Claude Desktop too. If you'd rather keep a local config, the mcp-remote bridge under Developer → Local MCP servers also works:

claude_desktop_config.json
{
  "mcpServers": {
    "sorvek": {
      "command": "npx",
      "args": [
        "mcp-remote",
        "https://app.sorvek.com/mcp",
        "--header",
        "Authorization:${AUTH_HEADER}"
      ],
      "env": {
        "AUTH_HEADER": "Bearer skv_mcp_your_key"
      }
    }
  }
}

Never add a url: entry in claude_desktop_config.json: Claude Desktop does not support that field and may silently rewrite the file, dropping the whole mcpServers section. The Authorization:${AUTH_HEADER} form (no space around the :) works around a known bug in Claude Desktop Windows and Cursor.

The 13 available tools

get_overviewKPIs: total apps/accounts, new apps in 30 days, AI apps, top users, last scan.
list_shadow_appsPaginated shadow IT inventory (search, category, risk, approval, date).
get_appFull app record: metadata, accounts, linked findings.
list_findingsSecurity findings from the latest scan (severity filter).
list_usersEmployees ranked by SaaS footprint (MFA, admin roles, risk).
get_user_footprintAn employee's full SaaS footprint (by email).
list_oauth_integrationsOAuth integrations aggregated by app, risk score and scopes.
get_shadow_ai_summaryScored Shadow AI summary (OAuth + browser extension, DLP leaks).
list_ai_agentsAI agent inventory (Copilot Studio + OAuth heuristics).
get_security_posturePosture rollup: SSO, MFA, Conditional Access, breaches, findings.
list_external_assetsExternal surface: subdomains, takeover risk.
list_recent_eventsActivity log: new apps, sign-ups, scans.
list_breach_alertsVendor breach alerts cross-referenced with your apps.

Example questions: "Which unapproved AI apps showed up this month?" · "Which at-risk employees hold sensitive OAuth grants?" · "Are there any subdomains at risk of takeover?"

Limits & security

60 calls / minute per key
Beyond that, the assistant receives an explicit message and retries on its own.
Scoped to your organization
Each key is tied to your tenant: the tools only see your data, from the active data source.
Immediate revocation
From Settings → API keys: the key dies on the next request. Every call is logged (tool, status, duration); personal parameters are redacted from the log.
Personal information
Some tools (list_users, get_user_footprint) return employee data (emails, departments, MFA status). This data flows to the AI provider you configure, so make sure that use respects your internal policies and obligations (Law 25 / GDPR).
Prompt injection
Returned content (app names, discovered domains) is observational data: Sorvek servers never emit instructions in their responses, and we recommend assistants with prompt-injection protections.

Sorvek Bridge is included in SaaS Discovery.

Not a customer yet? Run your first discovery in minutes.