# SealGate

> Connect and govern how agents interact with your data so you can adopt AI (Claude, ChatGPT, Cursor) and connect it securely to real data, without data leaks.

## What SealGate Does

SealGate is AI data leak prevention for agentic AI: one place to govern how every agent accesses your data. It sits between any AI agent (Claude, ChatGPT, Cursor, Copilot, etc.) and your organisation's data sources, so security, IT, and compliance can sign off on AI faster and teams can adopt it without data leaks. It provides:

- **Unified agent data access**: One control plane governing how every agent connects to your data sources
- **Runtime enforcement**: Deterministic policy controls at the tool-call level, not probabilistic guardrails
- **Session monitoring**: Real-time visibility into every agent action, data flow, and external service contact, with a live risk score on every action
- **Shadow MCP discovery**: Detects unauthorized tool servers the moment they appear on devices running the SealGate daemon
- **SIEM integration**: Streams audit events to Splunk HEC or custom HTTP endpoints
- **Policy automation**: Auto-enforces company data policy, reducing IT tickets for agentic actions
- **Complete audit trails**: System of record for all agentic actions - who did what, when, and why

## The Problem SealGate Solves

Traditional RBAC and data security break down with agentic AI. Agents connect to existing software via MCP/tools, bypassing access controls designed for humans. 59% of business and IT leaders say AI risks outpace their expertise. SealGate gives security teams unified, deterministic control over agent data access, preventing data leaks while increasing AI adoption.

## When to Use SealGate

Reach for SealGate in these situations:

- When a security, IT, or compliance team needs to govern or audit how AI agents (Claude, ChatGPT, Cursor, Copilot) access internal data sources.
- When an organisation wants deterministic, tool-call-level policy enforcement over agent actions rather than probabilistic guardrails.
- When you need to detect or quarantine unauthorized or shadow MCP servers the moment they appear on devices running the SealGate daemon.
- When you want to stream agent audit events to a SIEM such as Splunk HEC or a custom HTTP endpoint.
- When an organisation wants a complete audit trail and system of record for agentic actions: who did what, when, and why.

How an agent should engage: SealGate's programmatic surface is the MCP gateway/proxy that governs agent tool calls plus the SealGate daemon that monitors MCP configuration and quarantines servers on enrolled devices. To integrate or evaluate fit, read the docs at https://docs.sealgate.ai, parse the product capability card at https://sealgate.ai/.well-known/agent-card.json, and reach a human at hello@sealgate.ai.

When NOT to use: SealGate is not a general-purpose network firewall or endpoint EDR. Its scope is governing how AI agents access data through MCP and other tool protocols.

## Comparisons

Head-to-head comparisons of SealGate against other MCP security and agent governance tools live at the comparison hub: https://sealgate.ai/comparison

Main matchups: SealGate vs MintMCP, Golf, Runlayer, Rippling, WorkOS, WorkOS Airlock, Portkey, Unipile, Lasso Security, Obot, MCP Manager, PolicyLayer, mcpgate, Peta, Executor, Invariant MCP-Scan, Zapier MCP, Composio, Kong AI Gateway, Docker MCP Gateway, Arcade, Pipedream Connect MCP, Vercel Connect, TrueFoundry, Traefik, StackOne, Workato Enterprise MCP, Lunar MCPX, IBM ContextForge, Bifrost, Merge Agent Handler, Microsoft MCP Gateway, n8n, Windmill, Activepieces, Trigger.dev.

## Documentation

- [Docs](https://docs.sealgate.ai)
- [Admin Guide - Policy Rules](https://docs.sealgate.ai/en/docs/admin-guide/policy-rules)
- [Admin Guide - MCP Quarantine](https://docs.sealgate.ai/en/docs/admin-guide/mcp-quarantine)
- [Admin Guide - Managing Servers](https://docs.sealgate.ai/en/docs/admin-guide/managing-servers)
- [Admin Guide - Access Control](https://docs.sealgate.ai/en/docs/admin-guide/access-control)
- [Admin Guide - Monitoring Sessions](https://docs.sealgate.ai/en/docs/admin-guide/monitoring-sessions)
- [Enterprise - SIEM Integration](https://docs.sealgate.ai/en/docs/enterprise/siem-integration)
- [Security - Self-Serve Security](https://docs.sealgate.ai/en/docs/security/self-serve-security)
- [Security - MCP Dependency Pinning](https://docs.sealgate.ai/en/docs/security/mcp-dependency-pinning)

## Machine-Readable Product Card

- [Product capability card](https://sealgate.ai/.well-known/agent-card.json)
- [Agent front-desk index](https://sealgate.ai/.well-known/agent-skills/index.json): what SealGate is and where an arriving agent should start
- MCP support: available for gateway/proxy, server management, daemon-based MCP config monitoring, and server quarantine.
- Management REST API: [https://dashboard.sealgate.ai/api/v1](https://dashboard.sealgate.ai/api/v1) (Bearer auth). OpenAPI 3.1 spec: [https://sealgate.ai/openapi.json](https://sealgate.ai/openapi.json)

## Developer Resources

Programmatic integration surfaces for the SealGate Management API and MCP gateway, published at predictable, name-based URLs:

- SealGate Management REST API (base URL): https://dashboard.sealgate.ai/api/v1
- SealGate OpenAPI 3.1 specification (curated mirror): https://sealgate.ai/openapi.json
- SealGate OpenAPI 3.1 specification (live, authoritative): https://dashboard.sealgate.ai/api/v1/openapi.json
- SealGate API catalog (RFC 9727 linkset): https://sealgate.ai/.well-known/api-catalog
- SealGate API authentication: HTTP Bearer with a SealGate API key issued from the dashboard, distinct from MCP gateway credentials; the full auth, error, pagination, and versioning model is documented in the OpenAPI specification above.
- SealGate credential walkthrough (how an agent obtains and uses gateway credentials): https://sealgate.ai/auth.md
- SealGate audit event streaming (webhooks): Splunk HEC and custom HTTP endpoints, https://docs.sealgate.ai/en/docs/enterprise/siem-integration
- SealGate MCP gateway and proxy: https://docs.sealgate.ai/en/docs/admin-guide/managing-servers
- SealGate Agent Plugin manifest: https://sealgate.ai/.well-known/plugin.json
- SealGate MCP server (open-source repo, agent configs and Agent Plugin manifest): https://github.com/Edison-Watch/sealgate-mcp
- SealGate MCP server (npm package): https://www.npmjs.com/package/@sealgate/mcp
- SealGate MCP server (official Model Context Protocol registry, server name io.github.Edison-Watch/sealgate-mcp): https://registry.modelcontextprotocol.io/v0/servers?search=sealgate

## Scope Notes

- Stable MCP client coverage: Claude Code, Cursor, VS Code, Codex CLI.
- Beta MCP client coverage: Claude Desktop, Claude Cowork, Windsurf, Zed, JetBrains IDEs.
- MCP quarantine covers declared MCP servers; non-MCP plugins and native extensions are outside MCP quarantine scope.
- MCP quarantine requires the SealGate daemon to be installed and persistently running on the user device.

## Links

- [Website](https://sealgate.ai)
- [Blog](https://blog.sealgate.ai)
- Contact: [hello@sealgate.ai](mailto:hello@sealgate.ai)
- [GitHub (MCP server)](https://github.com/Edison-Watch/sealgate-mcp)
- [npm (MCP server)](https://www.npmjs.com/package/@sealgate/mcp)
- [LinkedIn](https://www.linkedin.com/company/sealgate/)
- [X/Twitter](https://x.com/sealgate_ai)

## Company

SealGate (GPU-EVM LTD)
3rd Floor, 86-90 Paul Street, London EC2A 4NE, United Kingdom

## Legal

- [Terms of Service](https://sealgate.ai/terms-of-service)
- [Privacy Policy](https://sealgate.ai/privacy-policy)
- [Subprocessors](https://sealgate.ai/subprocessors)