Skip to main content

Glean MCP Gateway

The Model Context Protocol (MCP) is an open standard that lets AI assistants connect to external data and tools through one consistent interface—think of it as a "USB for AI applications." The Glean MCP Gateway is how a Glean MCP server brings in everything external. It stands for anything outside Glean's own platform tools, in three flavors:

  • Custom tools: Read and write tools your team builds and packages in Glean.
  • External MCP servers: Third-party MCP servers you proxy securely through Glean.
  • Data source tools: Read and write tools from your connected data sources.

Through one secure, Glean-governed endpoint, the gateway exposes these tools to every assistant your teams use like Claude, ChatGPT, Cursor, Gemini, Microsoft Copilot, Claude Code, and more while enforcing permissions, policy, and audit on every call. It exposes 2,000+ governed tools this way, and as new tools are added they become available through the Glean MCP tools automatically.

To set up external tools, see Set up external tools with MCP Gateway. To proxy a third-party MCP server, see Connect remote MCP servers. For general MCP server concepts—architecture, supported hosts, the MCP configurator, authentication, and pricing—see About Glean MCP server.

Why a gateway​

MCP servers are appearing inside organizations faster than IT can track them. Without a central gateway, four risks compound:

  • Unvetted sources: Many MCP servers come from public repositories rather than official vendors.
  • Credential exposure: A poorly built server can leak credentials from the machine it runs on.
  • Governance gaps: There is no central catalog, approval flow, or policy engine.
  • Observability gaps: Anyone can install any server, leaving IT with no inventory and no audit trail.

Routing every MCP call through Glean closes these gaps.

What you get​

Vetted sources

Reach read and write tools connected to Glean, all through one approved endpoint.

Security and governance

Glean Protect+ policies apply to external MCP tool calls. Prompt injection, malicious code, and misaligned behavior can be blocked.

Observability

An Insights dashboard shows who uses which MCP server, which tools are called, and how often.

Management and access control

Manage every MCP server from one console—roll out, update, and revoke access—and limit tool calls to specific departments, roles, or data-sensitivity tiers.

How it works​

Glean MCP Gateway architecture: external hosts (Cursor, Claude Code, ChatGPT, VS Code) connect through one Glean MCP server endpoint to native tools, data-source-backed tools, custom tools, and other MCP servers.

One MCP endpoint exposes thousands of tools across native, data source, and custom tools to external AI hosts.

Each MCP server you create has its own URL (https://<tenant-id>-be.glean.com/mcp/<path>) that any supported host connects to. A single endpoint can serve Glean's own platform tools alongside everything the gateway brings in externally:

  • Glean platform tools: Search, chat, read document, code search, people, and more. Native to Glean, not part of the gateway.
  • Agents: Glean agents surfaced as tools that hosts can invoke. Native to Glean, not part of the gateway.
  • Custom tools (gateway): Read and write tools that admins build and package in Glean, such as internal APIs, CRM, ticketing, and workflows. See Set up external tools with MCP Gateway.
  • Data source tools (gateway): Read and write tools from your connected data sources, such as Jira, Salesforce, Slack, and Google, that respect source permissions.
  • External MCP servers (gateway): Third-party MCP servers such as Atlassian, GitHub, and Slack, proxied securely through Glean. See Connect remote MCP servers.

New servers typically start with core tools such as search, chat, read document, and more. For the full set of built-in tools and how to select them per server, see Create MCP servers.

note

We recommend including a maximum of 40 tools per MCP server. Plan which tools to expose per server accordingly.

All endpoints point to Glean, and Glean enforces what is allowed downstream. Because tools resolve against Glean's permission-aware Knowledge Graph, results stay unified and access-controlled across every connected source—unlike siloed native MCP servers.

Centralized management​

Because every endpoint routes through Glean, you manage MCP access from one place instead of configuring each host separately:

  • One console for every app: Register an MCP server once and distribute it to every supported host—Claude, ChatGPT, Cursor, Gemini, Microsoft Copilot, and Claude Code—from a single catalog of approved tools and servers.
  • Roll out, update, or revoke centrally: Enable or disable any server or tool for your organization from one place, without touching individual user configurations.
  • Fast rollout at scale: Glean provides a script to deploy MCP servers across managed devices in minutes. See Deploy MCP servers with MDM.
  • Control tool access per user: Decide who can use which tools from the admin console. By default, all tools are on for all users.
  • Permissions inherited from the source: Tools inherit source-system permissions, so Glean never exposes more than a user already has access to.
note

Tool visibility is governed by MCP access control, so each user sees only the tools they have access to—even when they connect to the same MCP server. For example, if a server is created with 5 tools, a user granted access to all 5 sees 5 tools, while a user granted access to only 2 sees just 2.

Common use cases​

One server for every tool

Create one server that includes Glean built-in tools, custom read and write tools, and external MCP servers—all in one place.

Developer workflows

Use Glean context in IDEs (for example, Cursor and VS Code) for debugging, PR reviews, and code navigation, and then close out your Jira tickets—all using one Glean MCP server.

Get work done

View your recent activity and use write tools to get work done through a single server.

Authentication​

The gateway uses the same authentication methods as any Glean MCP server: the Glean OAuth authorization server (recommended) or Glean API tokens. See Authentication for the full setup.

If a user invokes a downstream tool that needs its own credentials, the gateway prompts for that authorization and retries the call once complete.

Security​

Glean Protect+ guardrails can run on MCP calls and external tool calls—the same guardrails that protect Glean Assistant sit in front of the tool calls your users and agents make. Glean Protect+ can provide:

  • Prompt injection detection: Blocks adversarial inputs that try to hijack the agent.
  • Malicious code detection: Inspects tool inputs and outputs for known attack patterns.
  • Agent alignment checks: Verifies the tool call matches what the user asked for.
  • Restricted-topic policies: Per-group rules across users, data sources, and tools.

For session management, privacy, and data protection details, see About Glean MCP server. For a detailed security overview covering deployment, authentication, authorization, and monitoring, see MCP security.

Observability​

MCP Insights shows every call across every app:

  • Adoption: MAU, WAU, DAU, active-user trends, and usage among Glean users.
  • Usage breakdowns: active users by top host applications, users, applications, MCP tools, and MCP servers.

See MCP Insights for more information.

See also​

Was this page helpful?
Last updated on
© 2026, Glean Technologies, Inc.