About the IntelligenceBank MCP Server (BETA)

Modified on Wed, 7 Oct at 11:21 PM

The IntelligenceBank MCP Server connects your AI assistant to your IntelligenceBank platform, so you can search your assets and collections, read their details, see them in a gallery, download or resize them, upload a file, and run a risk review by asking for it in plain language. It is hosted by IntelligenceBank, there is nothing to install, and everything the assistant does runs as you, with your own permissions.

What the MCP Server Does

MCP, the Model Context Protocol, is an open standard for connecting AI assistants to the systems that hold your work. An MCP server publishes a set of tools, and the assistant chooses which to use to answer what you asked. The IntelligenceBank MCP Server publishes tools over your IntelligenceBank platform.

  • Hosted, not installed - the server runs on IntelligenceBank infrastructure. You point your AI assistant at a URL. There is no plugin, extension, or download.
  • Your platform, your permissions - you sign in to your own IntelligenceBank platform. The assistant sees exactly what you can see, and nothing else.
  • Ask, do not browse - there is no folder picker to click through. You describe what you want and the assistant runs the searches, follows the folder tree, and reads the metadata for you.
  • Speaks your organization's language - the assistant reads your platform's own module names, filters, and custom fields, so it uses the words your team uses rather than generic ones.
  • Built on an open standard - MCP is not a single vendor's plugin format, so the same server can serve any assistant that speaks it. During the beta we enable one AI assistant at a time, so tell us which one your team uses.

What You Can Ask For

You never call a tool by name: you ask for what you want and the assistant picks. The capabilities are summarised here because they define the boundary of what the assistant can and cannot do on your platform. The tools themselves, by name, are listed on the profiles page described in Profiles below. For example prompts for each capability, see Using the IntelligenceBank MCP Server: Use Cases and Example Prompts (BETA).

Capability What it does
Understand your platform Reads your organization name, the signed-in user, your module terminology, your brand colours and fonts, and which optional capabilities your site has switched on. The assistant calls this first so the rest of the conversation uses your names for things.
Search and browse Keyword search, structured filters using your own platform filters and custom fields, file type filters, and folder navigation. It can also find a folder by name, and look up a single asset from the short id at the end of a share link.
Find a collectionFinds your collections, the lightboxes and curated sets of assets you and your colleagues keep, by name, by owner, by who they are shared with, or by when they were created or changed. It can also tell you who each collection is shared with and what you may do with it. The assets in a collection are then searched like any others, so you can ask for "the PDFs in my Q3 campaign collection". Only the assets you are allowed to open come back. It is read only: nothing creates a collection or changes what is in it.
Read an asset Full metadata for one asset, including custom fields, ownership and usage terms, dates, and AI tags. Or download the original, a preview image, or a PDF rendering of an office document.
Show assets in a gallery Shows up to 20 assets as a gallery of their pictures, inside the conversation. Each picture links to the asset's page on your platform, to a download, and to its public link where public sharing is set up. The pictures load in your browser straight from your platform. An assistant that cannot display a gallery gives you the same links as text.
Resize or convert an image Produce a different size, crop, or format of an image, or apply one of your platform's Global Presets. IntelligenceBank does the rendering, so nothing is re-encoded along the way. Every image already carries three ready-made renditions, and the assistant is told to use one of those whenever it will do, because a transformation is a paid render on your platform.
Find a person Resolves a name or email address to an IntelligenceBank user, which is what makes questions like "everything Sam uploaded last month" answerable. It is lookup only: nothing creates, edits, or suspends an account.
Upload a file Sends a file from your AI assistant to IntelligenceBank, ready for a risk review. See Working With Files below for how the bytes travel.
Run a risk review Runs your organization's compliance, legal, and brand checks over an uploaded file, or over a piece of copy with no file behind it, and returns the findings. Requires the Marketing Compliance capability on your site. A file is always the better source, because it also runs the visual rules and places each finding on a page.
Renew your session Signs you back in to IntelligenceBank from inside the conversation when your session lapses, without going back to your assistant's connection settings.
Note: There is nothing here that edits or deletes an existing record. Metadata updates and approval requests are designed but not released, so today the only thing the assistant can add to your platform is a file you asked it to upload.

Profiles

A profile is a set of tools with its own server address. The address you connect with decides which tools your assistant gets. It can call only the tools its profile includes, so a narrower profile is a real limit on what that connection can do.

The default. The address https://mcp.intelligencebank.ai/mcp serves the default profile: every tool offered to customers, which is everything the other profiles offer, together. Start there unless you want to limit what a connection can do.

The profiles available. Open mcp.intelligencebank.ai in your browser. The page lists every profile, what it is for, and how many tools it has. Each profile's page lists its tools, grouped into those that only read and those that create something, with a button that copies the profile's address. Opening a profile's address in a browser shows the same page. Profiles change as the server grows, so the page is the current list, not this article.

  • Your permissions still apply - whichever profile you choose, the assistant reaches only what your IntelligenceBank account can reach.
  • Chosen when you connect - the profile is part of the address, so nothing in the conversation can change it. To use another profile, add another connection with its address.
  • Every profile can renew your session - each one includes the tools that read your platform's settings and sign you back in, so a narrow profile never leaves you stuck after a session lapses.

Requirements

  • An active IntelligenceBank account, on a platform that has been enabled for the MCP Server beta.
  • An AI assistant that supports remote MCP servers over HTTP. Assistants that only support MCP servers installed on your own machine cannot reach this one.
  • Support for OAuth 2.0 sign-in in that assistant. There is no API key to paste: authentication is a browser sign-in to your own platform.
  • Permission to add a custom MCP server or connector, or an administrator who adds it for you. Some AI assistants restrict this by plan, or let a workspace administrator control it centrally.
  • An internet connection. Every search, download, upload, and review is an online operation.

Connecting Your AI Assistant

Every AI assistant words this differently, and their screens change often, so this article gives you the values to enter and links to each assistant's own instructions rather than repeating its click path. Use these values wherever your assistant adds a remote MCP server, a custom connector, an app, or a tool.

Server URL https://mcp.intelligencebank.ai/mcp for the default profile, or the address of another profile from mcp.intelligencebank.ai.
Transport HTTP. Choose the HTTP or "remote server" option, never the local or command-line one.
Authentication OAuth 2.0. If your assistant asks whether authentication is required, the answer is yes, always. See the warning below.
OAuth client ID Leave empty where your assistant uses its own published identity, as Claude does, or registers itself, as Microsoft Copilot Studio does. If it asks for a client ID, enter mcp-public-client.
OAuth client secret None. Leave it blank.
Platform URL Not part of the connection. You enter it during sign-in instead, which is what tells the server which IntelligenceBank platform you work on. Typically, you will see an action to Connect after you configured the previous fields.
Warning: Some AI assistants may probe the server and conclude that no authentication is needed. That detection is wrong, and it fails quietly. Choose the option that always signs in, or the one that signs in when the server asks.
Tip: If your assistant offers to use its own published identity, or to register itself automatically, choose that and leave the client ID and secret empty. Enter mcp-public-client only when your assistant asks you for a client ID.

Administrator or Individual User

Most AI assistants offer two ways to add a server. An administrator adds it once for the whole organization and decides who can use it. Or each person adds it for themselves, where their plan and their administrator allow it. Either way, every person signs in to IntelligenceBank with their own account the first time they use it, so no credential is ever shared.

The links below go to each assistant's own documentation, which is always current for its screens. What is specific to IntelligenceBank is in each table.

Claude (web, desktop and Cowork)

Who adds it On Team and Enterprise plans, an Owner adds it for the organization, and each member then connects to it and signs in. On Free, Pro and Max plans, you add it yourself as a custom connector.
What to choose The server URL above. For the OAuth client, keep Use Claude's published identity. Nothing is pasted.
Tool controls Owners on Team and Enterprise plans can limit which tools a connector may use across the organization. Each person can also switch tools off for a conversation, and choose which tools Claude may run without asking first.
Instructions Get started with custom connectors using remote MCP
Use connectors to extend Claude's capabilities

Claude Code

Who adds it Each developer, with one command. An administrator can also deploy servers to everyone in the organization, or restrict which servers are allowed.
What to choose Run claude mcp add --transport http intelligencebank https://mcp.intelligencebank.ai/mcp, then run /mcp inside Claude Code to sign in. No client ID is needed.
Tool controls Permission rules allow a tool, ask before running it, or deny it, per tool.
Instructions Connect Claude Code to tools via MCP
Control MCP server access for your organization

Microsoft Copilot Studio

Who adds it The person who builds the agent adds the server to it as a tool. Everyone who uses the agent then signs in with their own IntelligenceBank account the first time. Administrators govern MCP servers through Power Platform data policies.
What to choose The server URL above. Authentication OAuth 2.0, type Dynamic discovery. Nothing else to fill in: Copilot Studio registers itself with our sign-in service.
Tool controls Every tool is on when the server is added. The agent's builder can switch individual tools off, and once they do, tools added to the server later start switched off.
Instructions Connect your agent to an existing Model Context Protocol (MCP) server
Add tools and resources from an MCP server to your agent

Controlling Which Tools Are Enabled

Two things decide which tools your assistant can use, and an administrator sets both while adding the server.

  1. The profile in the server address. A connection can call only the tools its profile includes. IntelligenceBank enforces this, whatever the assistant does. An administrator who adds the server for the organization chooses the profile for everyone using that connection.
  2. Your AI assistant's own controls. Most let an administrator switch individual tools off, or require approval before a tool runs, and many let each person do the same for themselves. What each assistant offers is in its table above, with a link to its documentation.

Every tool also tells your AI assistant whether it only reads or creates something. Assistants use that to decide which tools to ask you about, and the profiles page shows the same grouping. On top of all of this, your IntelligenceBank permissions apply: no tool reaches anything your account cannot.

Signing In and Sessions

1Connect Once the connection details are saved, your assistant shows a Connect, Authorize, or Sign in action on the connection. Click it. Saving the details alone does not sign you in. Some assistants skip this and start the sign-in the first time you ask for something from IntelligenceBank instead.
2Name your platform A browser window opens on the IntelligenceBank sign-in page and asks for your platform URL, for example yourcompany.intelligencebank.com.
3Sign in Sign in the way you normally do, single sign-on included. The window returns you to your assistant when it is done.
4Carry on The connection now shows as connected and the tools work from then on. Nothing else has to be configured, and if the sign-in started from something you asked for, that request is retried for you.

This is where your platform is chosen. The platform URL you enter during this sign-in is what binds the connection to your IntelligenceBank site, and it is the only place it is ever asked for. Nothing you type in the conversation afterwards can change it.

How long a session lasts. Your IntelligenceBank session stays alive for as long as your platform's login timeout allows, between 1 and 120 hours, set by your administrator. When it lapses, the assistant hands you a sign-in link in the conversation and picks up where it left off once you have used it. That renewal signs you back in to the same platform, so you should not need to remove and re-add the connection to recover an expired session, and if your IntelligenceBank browser session is still alive it is usually a single click.

One platform per connection. To work on a different IntelligenceBank platform, remove the connection and add it again, signing in to the other platform when it asks. Switching is a connection-level action, never something the assistant can do mid-conversation.

Working With Files

Uploads go direct. File contents never travel through the conversation. When you upload, your assistant sends the bytes straight to IntelligenceBank's storage, and only a fingerprint of the file passes through the server. Assistants that run code in a sandbox need outbound network access for this. If the sandbox is blocked, the tool tells you the exact address to allow.

Downloads come straight from your platform. When you ask for the file itself, your assistant receives a download link on your own IntelligenceBank platform, and the file goes from there to your assistant without passing through the MCP Server. The link works while your session is active. What happens next depends on your assistant: some open the link and can work with the file, others show it to you to open yourself.

Reviews take minutes, not seconds. A risk review runs for one to five minutes, which is longer than most AI assistants allow a single step to take. The assistant is told to check back repeatedly rather than wait, so expect a few short pauses before the findings arrive. That is normal and not an error.

Warning: A public image link bypasses IntelligenceBank permissions. Anyone who has the link can open the image, whether or not they have an IntelligenceBank account. Public sharing also has to be configured on your platform and on the folder of the asset for a link to work at all.

Data and Security

  • Encrypted in transit - all communication with IntelligenceBank uses TLS.
  • Your authentication - sign-in is delegated to your IntelligenceBank platform, including SSO if your organization has it configured. The MCP Server never sees or stores your password.
  • Same platform controls - every call is made as you. Your existing permissions, retention policies, and access controls apply exactly as they do in the web application, and a file you upload lands under them too.
  • Nothing is warehoused - the server holds your session and nothing else. It does not copy your files or the data returned.
  • Your assistant sees what it retrieves - results come back into your AI assistant's conversation, so your AI provider's own data handling terms apply to them. That is worth checking against your organization's policy before you connect a platform holding sensitive material.

Frequently Asked Questions

Which AI assistants does this work with?

Any AI assistant that supports remote MCP servers with OAuth sign-in can connect, because MCP is an open standard. Claude, Claude Code and Microsoft Copilot Studio are the ones we have tested, and Connecting Your AI Assistant covers their setup. With another assistant, whether the sign-in completes depends on how that assistant identifies itself to our sign-in service. Tell us which one your team uses, because that decides which we test next.

Which profile should I use?

Start with the default, https://mcp.intelligencebank.ai/mcp. Choose a narrower profile when you want a connection that cannot do certain things, for example one that never uploads a file. mcp.intelligencebank.ai lists what each profile includes.

The assistant says it cannot do something this article describes.

Your connection may use a profile that does not include that tool, or an administrator may have switched the tool off in your AI assistant. Compare the connection's address with the profiles page, and ask your administrator about the assistant's tool settings.

Do I need an API key or a service account?

No. You sign in as yourself, in a browser, on your own platform. There is nothing to generate, paste, or rotate, and no shared credential for your team to look after.

Can the assistant change or delete our assets?

No. The published tools read, search, download, transform images, upload a file, and run a risk review. There is no tool that edits metadata, moves a record, or deletes anything. Everything it can do is also bounded by your own IntelligenceBank permissions.

Why does my assistant say the server needs no authentication?

Because the questions it asks while probing are answered without a credential by design. Discovery has to succeed for the connection to survive, so the server never refuses those. Only a real tool call triggers the sign-in. Set authentication yourself rather than trusting the detection, as described above.

A new capability was announced but my assistant does not show it.

Some AI assistants cache the list of tools a server publishes, and refresh it on their own schedule, which can be up to a day. Nothing is broken in the meantime and the tools you already have keep working. Disconnecting and reconnecting usually does not shorten the wait. In some assistants an administrator has to refresh the app before new or changed tools are used. The assistant's own documentation says whether yours does.

Nothing is working. What should I check?

  1. Confirm the server URL is exactly https://mcp.intelligencebank.ai/mcp, or a profile address copied from the profiles page, including the path at the end.
  2. Confirm authentication is set to required, not none.
  3. If you entered an OAuth client ID yourself, confirm it is mcp-public-client with an empty secret.
  4. Ask the assistant something that needs IntelligenceBank, so the sign-in is actually triggered. A connection that lists tools has not necessarily signed you in.
  5. Check you can sign in to your platform in a normal browser tab.
  6. If it still fails, submit a ticket and tell us which AI assistant you are using and what it showed you.

Need Help?

To request access for your site, contact IntelligenceBank, and tell us which AI assistant your team uses. For anything else about the MCP Server that needs our team, you can submit a ticket.

Beta and Early Access

Our IntelligenceBank MCP Server is currently in BETA. Contact us to request early access.

While it is in beta, the set of tools and the way they behave can change between releases, and your feedback shapes what comes next. Tell us what you tried, which AI assistant you used, and what you wanted it to do.

Request Early Access

Was this article helpful?

That’s Great!

Thank you for your feedback

Sorry! We couldn't be helpful

Thank you for your feedback

Let us know how can we improve this article!

Select at least one of the reasons
CAPTCHA verification is required.

Feedback sent

We appreciate your effort and will try to fix the article