Secure Your Data With Grok

MCP is native. xAI shipped remote MCP support on the Grok API and added 'Bring Your Own MCP' in 2026, alongside the Grok Connectors launch (May 6, 2026). PortEden registers as a first-class MCP server, and Grok calls it the same way it calls any other MCP endpoint. Function-calling is still supported if you'd rather keep that path — PortEden CLI commands map 1:1 to Grok function declarations.

Native Grok Connectors ground on the user's raw OAuth scope, which is the same oversharing risk M365 Copilot hit on its rollout. PortEden's proxy applies per-site, per-contact, and per-label rules in front of those connectors, so Grok grounds on the data your policies have explicitly approved — not just on whatever the user can technically access.

Yes — model choice is on the Grok side. PortEden returns the same redacted JSON regardless of which Grok variant calls it. Use Grok-3 for cheap high-volume tool calls and Grok-4 when you need extended reasoning loops or deeper analysis.

xAI sees the redacted output PortEden returns to the tool call — never the raw email, calendar, or drive payload. Identifiers, free-text PII, and any field your access rules block are replaced with placeholders before the response leaves PortEden's perimeter. PortEden audits the tool call request and the redacted response; we don't see the user's prompt to Grok or Grok's final answer.

The CLI or REST endpoint returns a structured permission-denied response. Grok sees the rejection and surfaces it in the answer ("I tried to read that file but your firewall blocked it"). The audit log captures the tool call, the arguments, and the rule that fired — no silent failures.

Yes. Pass a per-user PortEden token as a Bearer header on every tool call. Each call is audited under the calling user's token, and access rules apply per-token — so the same Grok deployment can serve multiple users without leaking data across them.

Every tool call in the loop routes through PortEden, so redaction and RBAC apply on every step. Per-token monthly quotas keep loops bounded — exceed the quota and the next tool call gets a clean "quota exceeded" response, which Grok can surface to the user. The audit log captures the full chain, end-to-end.

Redaction adds ~150-200 ms per tool call. Because PortEden also reduces token usage by ~80% via context hygiene, end-to-end latency typically drops on data-heavy tasks. Grok answers faster on PortEden's clean payloads than on a raw API response.

Yes — free for solo licensed practitioners and small developer use. Higher monthly tool-call quotas, SSO, and SIEM export are on paid plans. See pricing for details.