Glean MCP Server

Glean is an enterprise search and AI assistant platform that indexes content across an organization's connected SaaS apps (Google Drive, Slack, Confluence, Jira, GitHub, Salesforce, and many more). The Glean MCP server exposes that unified Knowledge Graph to any MCP-compatible client (Claude Desktop, Cursor, VS Code, Windsurf, Goose, JetBrains AI, Gemini CLI, etc.) so an AI agent can search company content, chat with Glean Assistant, look up people, and pull full document contents, all while enforcing each user's existing access permissions.

Glean offers two deployment modes. The remote MCP server is hosted by Glean as part of every Glean instance and is the recommended option: admins enable it in the Admin Console under Platform → Glean MCP server, users authenticate via OAuth 2.1 with PKCE, and a built-in MCP Configurator generates client-specific setup snippets. The local MCP server is an open-source TypeScript implementation in the gleanwork/mcp-server monorepo, runnable via npx or Docker and authenticated with a Glean API token; it is positioned as a fallback for experimentation.

Both servers expose the same four core tools: company_search, chat, people_profile_search, and read_documents. Because results are scoped to each user's permissions in the underlying source systems, the server is safe to deploy across an organization without exposing data that the requesting user could not see directly in Glean.

Tools

Option A: Remote MCP server (recommended)

Glean hosts the MCP server inside your Glean instance. Authentication is OAuth 2.1 with PKCE; no API token management required for end users.

• An admin enables it once: Admin Console → Platform → Glean MCP server → Enable.
• Each user opens their Glean profile → Your Settings → Install → MCP Configurator and picks their client (Claude Desktop, Cursor, VS Code, Windsurf, Claude Code, Gemini CLI, Goose, JetBrains AI, OpenCode).
• The Configurator emits a tailored config snippet pointing at your instance's MCP URL (typically https://<instance>-be.glean.com/mcp) and walks through the OAuth handshake.

Option B: Local MCP server (open-source)

Self-hosted via npx or Docker, authenticated with a Glean API token.

Automated configuration:

npx @gleanwork/configure-mcp-server --client cursor \
  --token YOUR_GLEAN_API_TOKEN \
  --server-url https://your-instance-be.glean.com

Supported --client values include cursor, claude, and windsurf.

Manual configuration (Claude Desktop / Cursor / Windsurf):

{
  "mcpServers": {
    "glean": {
      "command": "npx",
      "args": ["-y", "@gleanwork/local-mcp-server"],
      "env": {
        "GLEAN_SERVER_URL": "https://your-instance-be.glean.com",
        "GLEAN_API_TOKEN": "your-glean-api-token"
      }
    }
  }
}

Docker:

{
  "mcpServers": {
    "glean": {
      "command": "docker",
      "args": ["run", "-i", "--rm", "ghcr.io/gleanwork/local-mcp-server:latest"],
      "env": {
        "GLEAN_SERVER_URL": "https://your-instance-be.glean.com",
        "GLEAN_API_TOKEN": "your-glean-api-token"
      }
    }
  }
}

Environment variables:

• GLEAN_SERVER_URL (recommended): your Glean backend URL, e.g. https://your-instance-be.glean.com.
• GLEAN_INSTANCE (deprecated): legacy instance identifier.
• GLEAN_API_TOKEN (required): API token generated in the Glean admin console.

Prerequisites

• A Glean workspace with the source apps already indexed.
• Admin access to enable the MCP server (remote) or to mint an API token (local).
• A Glean plan that includes API/MCP access.

• Ask an AI coding agent "what's our internal style guide for React components?" and have it pull the answer from Confluence, Notion, and GitHub wikis via Glean, with citations.
• During incident response, query Glean from Claude Desktop to surface relevant Slack threads, Jira tickets, and runbooks for a given service name.
• Have an agent find the right subject matter expert ("who owns the billing service?") using people_profile_search and draft an introduction message.
• Summarize a long internal RFC by passing its URL to read_documents, then chain with chat to compare it against related proposals.
• Build a sales prep workflow that pulls all recent activity (emails, Salesforce notes, Gong calls) about an account from Glean before a customer meeting.

• "Search Glean for our latest Q3 pricing deck and summarize the changes from Q2."
• "Use Glean chat to answer: what's our policy on open-source contributions?"
• "Find the engineering manager who owns the payments service and show their profile."
• "Read these two Glean documents and tell me which one is more recent and what they disagree on."
• "Search Glean for any tickets or Slack threads mentioning the 'checkout-timeout' bug from the last 30 days."

• Official and actively maintained by Glean, with both a managed remote server and an open-source local server.
• Permission-aware: results respect each user's existing access controls in the underlying source systems.
• Unified search across dozens of enterprise apps, far broader than any single-app MCP server.
• First-class support for major MCP clients (Claude Desktop, Cursor, VS Code, Windsurf, Gemini CLI, JetBrains AI, Goose, Claude Code).

• Requires a paid Glean subscription and a fully configured Glean instance with indexed data sources.
• The local MCP server is explicitly positioned as experimental; Glean recommends the remote server for production use.
• Tool surface is intentionally narrow (search, chat, people, read), no write actions like creating documents or tickets.

• Guru MCP server for company knowledge cards.
• Per-app MCP servers (Notion, Confluence, Slack, Google Drive) when you only need one source and don't have Glean.
• Self-hosted RAG stacks built on tools like LlamaIndex or Vectara for teams without Glean.