Can any employee publish a custom MCP server inside Glean?

The documented setup is administrator-controlled. Authorized end users invoke the selected tools after setup and complete their own authorization where required.

Does an administrator’s OAuth connection authorize all users?

Not when per-user OAuth is configured. Each end user must authorize separately. Check the chosen authentication model before treating a setup test as a user-access test.

Should I use Glean’s own MCP server to add outside tools to Glean?

No. That server exposes Glean to another assistant. To bring outside tools into Glean, an administrator configures the outside service as a remote MCP server.

What belongs in a connection handoff?

Save the connection direction, approved setup reference, selected tools, owner, test results, and next unresolved check. Do not include credentials or sensitive source content.

Glean MCP: connect knowledge and tools in the right direction

Decide whether another assistant should use Glean or Glean should use an outside tool. Then configure that direction, authorize the right users, and test a small request.

July 2, 2026Updated October 28, 20265 min read

Glean MCP has two directions: another assistant can use Glean’s MCP server, or Glean can use tools from an outside MCP server. Glean’s remote-server guide (opens in a new tab) explicitly distinguishes them. MCP is a standard way for AI apps to use outside tools; choose which app will make the requests before configuring anything. fenbs can keep the approved connection details and test findings so the next setup session starts from what you already established.

Write the intended connection as a sentence: “Glean Assistant should read this outside work record,” or “My other assistant should ask Glean for company information.” These are different jobs. Adding Glean’s own server to a different app does not publish an outside tool inside Glean.

Which connection direction do you need?

Your goal	Direction	Starting point
Use Glean from another AI app	The other app calls Glean’s MCP server.	Get your organization’s approved Glean server setup for that host.
Use an outside service from Glean	Glean calls the outside service’s MCP server.	An administrator configures and publishes the permitted remote tools.

Glean’s documentation directory (opens in a new tab) lists its MCP server resources separately from its tools administration resources. For the first direction, get the approved endpoint and host instructions from your Glean administrator. Do not derive a server address from your company’s web address or paste a guessed configuration into an assistant. This guide does not substitute an invented endpoint for organization-specific setup.

Before proceeding, identify the user who will authorize access, the information involved, and the smallest operation that would prove the connection works. If the only goal is to find and verify indexed content, Glean enterprise search covers that task without changing the tool setup.

Where does an administrator add outside tools to Glean?

Check the Connectors catalog first for catalog-backed integrations. Glean’s remote MCP setup guide (opens in a new tab) distinguishes that route from custom or unlisted servers and existing Platform → Tools configurations. Use the documented route for your integration instead of assuming one universal menu path. Existing connections are not automatically moved into the newer catalog flow.

For a custom or unlisted server, open Admin console → Platform → Tools → Vendor Provided Tools (via MCP).
Select Add, open Vendor provided tools (via MCP), then choose Import tools from MCP server.
Enter the server name, description, URL, transport, and authentication supplied for that service.
After connecting, use Enable tools → Edit settings and enable only the required tools for Glean and/or Agents.
Select Save, then test the published tools as the intended user.

Give the connection a description that explains its purpose. “Read the approved work record” is easier for a reviewer to understand than “MCP test.” Before publishing tools that change information, agree on the allowed actions and how a person will review their inputs. Start with a reading operation when that is enough to prove the intended access.

Which transport and authentication details matter?

The Glean connection guide (opens in a new tab) documents HTTPS endpoints, SSE and Streaming HTTP, and several authentication choices, including admin-shared and per-user OAuth. A private endpoint also needs the relevant network connectivity. Use the service’s supplied transport and authentication details rather than choosing an option because its name looks familiar.

OAuth means the user or administrator signs in to authorize access. The important question is whose authorization a request uses. With per-user OAuth, each user authorizes separately; the administrator’s initial discovery connection does not sign everyone else in. Shared authorization and individual authorization therefore need different access reviews.

Record the endpoint and transport without putting secrets in the setup record.
Name the authentication approach and the person responsible for maintaining it.
List the tools selected for publication and why each is needed.
Check whether the intended users must complete their own authorization before testing.

Keep passwords, access tokens, and private customer information out of a reusable checklist. Save where the approved instructions live and who owns the connection. Connecting AI apps becomes easier when the next person can distinguish a reusable setup fact from a credential they should never copy.

How do you test the connection in either direction?

Test one permitted reading request with information the user is allowed to see. State the expected outcome before running it. For example, choose an approved reference whose title and content you already know. Inspect whether the tool actually ran and whether its result supports the assistant’s final answer. A confident response alone does not demonstrate that the connection was used.

Confirm the intended user has completed the required authorization.
Request one narrow piece of approved information through the connected tool.
Compare the returned result with the known source and expected access.
Test an approved negative example that should be refused or unavailable.
Record the actual outcomes and any unresolved access question before widening use.

For an outside assistant calling Glean, use the host instructions your administrator provided. For Glean calling an outside service, test the published tool in Glean Assistant. Keep the two results separate: success in one direction says nothing about whether the other was configured. Save checked findings in fenbs’s work record so another assistant can continue the investigation.

Why might discovery work while a real request fails?

Separate the stages: reaching the endpoint, discovering tools, publishing them, authorizing the user, and calling the selected tool. Identify the first stage that fails instead of replacing the entire configuration. An administrator seeing tools during setup does not establish that an end user has authorized or can perform every operation.

When escalating, include the connection direction, selected tool, transport, authentication approach, and failure message with sensitive details removed. Ask whether the failure matches the expected permissions. A denied operation may be the correct outcome. Do not bypass it by sharing an administrator’s credentials.

Glean’s MCP guide (opens in a new tab) documents full remote MCP support in Assistant and beta support in agents. Published MCP tools work in Plan and execute steps and autonomous agents, rather than as individual single-step selections. If you are extending a workflow, building Glean agents covers the review points; do not assume Assistant success establishes identical agent behavior.

How fenbs helps

Never lose the work you did in an AI session

After a useful AI session, save it in fenbs: the key observations, findings, research and context. Then carry on from exactly where you left off, on any device and in any other AI. The next AI you connect reads it at the start of the session, so you never have to explain the same problem again.

Create a free account

Free for up to 5 people and 50 open tasks · No credit card · Works with Claude, ChatGPT, Cursor and more

How to keep a checked Glean connection with fenbs?

Keep the approved connection details, the results you checked, and the next unanswered question. You can then carry on from the same point on any device and in another connected AI without repeating the setup investigation.

Create a free account.
Add a task for the connection. Write which app calls which service, who owns it, and the information it is allowed to use.
Connect Glean to fenbs using Integrations, with your administrator configuring and publishing the permitted remote tools.
Save the approved setup reference, tool choices, and test results. Leave credentials in their approved secure location.
Save where you left off and the next check. Continue on any device with another connected AI without losing the setup work.

fenbs preserves the work between sessions; it does not act as a network proxy, publish Glean tools, or grant access to company information. If you only need one short test and no future handoff, a separate work record may be unnecessary. Keep it when another person or assistant will need to understand what was authorized and verified.