Skip to main content

How to use the ManyRequests MCP server

Connect Claude, ChatGPT or another AI app to your portal, let your team members and clients connect too, and see or revoke every connection.

Written by Gabriel Lecointere

Plan requirement: the MCP server is available on the Pro and Core plans. Portal admins and admins can always connect an AI app. Team members and clients can connect too once you allow it.

Use case: you want to manage your portal from your AI assistant: check requests, update statuses, create invoices and keep an eye on your team's workload, in plain language and without switching tabs. Your account managers want to catch up on their clients' requests from Claude, and your clients want to send a new request or check their hours left from ChatGPT. The MCP server connects Claude, ChatGPT or any MCP-compatible AI app to your portal, for each of them.

What is the MCP server?

MCP (Model Context Protocol) is an open standard that lets AI apps connect to other tools and take actions for you. Once you connect the MCP server, your AI app can read and change data in your portal when you ask it to.

Good to know: there are no tokens or API keys to copy. Your AI app asks you to log in to your portal and approve the connection. Everyone connects with their own login, and the AI only sees and does what that person can see and do in the portal.

Who can connect

Role

Can connect

Where they find the link

Portal admins and admins

Always

Settings > Integrations > MCP server, or the AI assistants tab of their profile

Team members

When Team members can connect AI assistants is on

The AI assistants tab of their profile

Clients

When Clients can connect AI assistants is on

The AI assistants tab of their profile. See How to connect your AI assistant to your portal as a client.

Both switches are off until you turn them on.

How to let your team or your clients connect

  1. Go to Settings > Integrations and open the MCP server tab.

  2. Scroll to Who can connect. It says: Admins can always connect. Turning a role off also cuts the connections that role already has.

  3. Turn on Team members can connect AI assistants, Clients can connect AI assistants, or both. The change is saved right away and you see MCP access updated.

    Who can connect card on the MCP server tab with both switches on and the MCP access updated confirmation

Once a role is allowed, everyone with that role gets an AI assistants tab in Settings > Profile and account, with the steps and the link to copy.

AI assistants tab of Profile and account for a team member, with the server link and Copy button

Good to know: turning a role off disconnects every AI app that role already connected, right away. Their AI assistants tab disappears, and their AI app gets the message This portal does not let your role connect an AI assistant. If you turn the role back on later, they need to connect again.

Where to find your server URL

  1. Go to Settings > Integrations and open the MCP server tab.

  2. Under Install, click Copy next to the server URL. It is your portal address followed by /mcp. If you set up a custom domain, the link uses it.

    MCP server tab with the Copy button for the server URL highlighted

The link is the same for everyone in your portal. Team members and clients copy it from the AI assistants tab of their profile.

Connect with the ManyRequests link or from the connector directory

There is also one link that works for every portal: https://app.manyrequests.com/mcp. With it, people don't need your portal's link in advance: they type your portal address when they connect. The same rules apply as with your own link: admins can always connect, and team members and clients can connect when you allow it.

If ManyRequests is listed in your AI app's connector directory, connecting from there works the same way: pick ManyRequests, click Connect, then follow steps 2 and 3 below.

  1. In Claude, go to Customize > Connectors, click Add custom connector, paste https://app.manyrequests.com/mcp, click Add, then click Connect.

  2. A ManyRequests page opens: Connect Claude to your portal. Under Portal URL, paste any link from your portal, for example your portal address, and click Continue.

  3. Log in to your portal and click Authorize on the approval screen (see What the approval screen shows below).

The next time you connect, the page offers Continue to your portal, or Use a different portal.

Good to know: with this link, the AI app lists the connector as ManyRequests, and the page where you type your portal address shows the ManyRequests logo. The login and approval screens after it are your portal's. If your clients shouldn't see ManyRequests, give them your portal's own link instead. For Cursor, VS Code and Claude Code, use your portal's own link.

How to connect Claude

The same steps work in the Claude desktop app and at claude.ai.

  1. In Claude, go to Settings, then Connectors, and click Add custom connector.

  2. Paste your server URL and click Add.

  3. Log in to your portal and approve the connection when Claude asks (see the next section).

  4. The first time Claude uses the server in a conversation, it asks for permission. Allow it once or always.

How to connect ChatGPT

The same steps work in the ChatGPT desktop app and at chatgpt.com. This requires a paid ChatGPT plan.

  1. Go to Settings, then Apps & Connectors, then Advanced settings, and turn on Developer mode.

  2. Back in Apps & Connectors, click Create, name the connector, paste your server URL and choose OAuth for authentication.

  3. Log in to your portal and approve the connection when ChatGPT asks.

  4. In a chat, open the + menu, choose Developer mode and enable the connector.

What the approval screen shows

After you log in, your portal shows an approval screen with your portal logo (or its name), so your team and clients see your brand, not ManyRequests. It shows:

  • Authorize Claude (or the name of the AI app) and Claude wants to access your Nexus Digital account.

  • Logged in as and the email of the account being connected.

  • Authorization will be sent to and the address of the AI app, for example claude.ai. Check it matches the app you are connecting.

Click Authorize to connect, or Cancel to stop.

Portal approval screen asking to authorize Claude for the Nexus Digital account

How to connect Cursor, VS Code or Claude Code

Under For developers on the MCP server tab:

  • Click Add to Cursor or Add to VS Code to install the server in that editor.

  • For Claude Code, click Copy and run the command in your terminal. It looks like claude mcp add --transport http --scope user your-portal https://your-portal-address/mcp.

For developers section with Add to Cursor, Add to VS Code and the Claude Code command

These buttons are only on the MCP server tab. A team member who works in one of these editors can add the server URL from their AI assistants tab in the editor's MCP settings.

Check that the connection works

Start a new chat and paste one of the prompts from Try these prompts on the MCP server tab.

Try these prompts section with two example prompts

Start each conversation with a workspace check

Ask "Which portal am I connected to?" first. The AI loads your portal name, statuses, tags and team members, which it needs to turn names into the right records when it creates or updates requests. Once per conversation is enough.

What you can do with the MCP server

The What you get section on the MCP server tab lists every tool the MCP server has. Each person only gets the tools their role and permissions allow (see below). Here is what they cover, with example prompts.

Requests

Find, read, create, update, complete and comment on requests.

  • "Which requests are overdue?"

  • "Find all requests assigned to Sam."

  • "How many open requests does Northwind Studio have?"

  • "Read the conversation on request #1007 before I reply to the client."

  • "List the files we delivered on request #1007."

  • "Create a Design Subscription request for Hannah Brooks titled Spring newsletter header, due next Friday."

  • "Move request #1012 to In progress and assign it to Elena."

  • "Change the priority of the trail map request to high."

  • "Mark request #1012 as completed."

  • "Leave an internal note on request #1007: client approved the first draft, go ahead with the final version."

When you update a request, the AI replies with a link so you can open it in the portal. Priority can be low, medium or high. The AI can read the files on a request and the files posted in its comments, but files are still uploaded in the portal.

Marking a request as completed moves it to your completed status and notifies everyone involved, as the Mark as completed button does. A request that is already completed is left as it is.

Clients and organizations

  • "Show me all the clients in Bluebird Bakery."

  • "What does Hannah's account look like: what's open and what is Northwind Studio subscribed to?"

  • "Create an organization called Pinecrest Studio with owner [email protected], and don't send the invitation email yet."

  • "Fix the spelling of the client name: it should be Leo Martins."

Services and subscriptions

  • "How much does the Design Subscription cost?"

  • "What fields does the Landing Page request form ask for?"

  • "Subscribe Copperleaf Coffee to the Content Retainer, monthly."

  • "Add one Website Audit to Harbor & Pine Legal."

Adding a subscription or a one-off service through the MCP server never charges the client. Bill it separately with an invoice if needed.

Invoices

  • "What invoices do we have for Northwind Studio this month?"

  • "Create an invoice for Maria Gonzalez for one Ad Creative Pack."

  • "Send a payment reminder to Grace Liu for her pending invoice."

  • "Mark the invoice for the website audit as paid by bank transfer."

What each role can do

The AI works as the person who connected it. It gets the tools that match their role, and the role permissions in Settings > Modules and extensions > Users apply to it as they do in the portal.

Role

What the AI can do

Portal admins and admins

Everything above. The invoice tools need invoice management turned on, and admins also need the Invoices permission.

Team members

Find and read requests, update them, mark them as completed, comment, look up services and read the request activity. With Create requests, they can also create requests for clients. With Access all clients or Access managed clients, they can also look up and edit clients and organizations (creating or renaming organizations needs Access all clients).

Clients

Send new requests, follow their requests, read and download the files delivered, comment, mark a request as completed, check their plans and the hours or credits left, and read their invoices. See How to connect your AI assistant to your portal as a client.

A few examples of how permissions carry over:

  • A team member without Access all requests only reaches the requests assigned to them.

  • Without Access request activity, the AI can't read or post comments, or read the activity.

  • If someone can't see the priority, tags or due date of a request in the portal, the AI doesn't get them either.

  • Renaming a request, or changing its status, priority, tags, due date or assignees, needs the same permission as in the portal. If one change isn't allowed, nothing in that request is changed.

Ideas for agencies

Workload: "Which team member has the most open requests right now?", "Show me the workload for each team member this week as a chart."

Capacity planning: "What's due for each team member next week?", "Who has the most requests due in the next seven days?"

Client health: "Which clients haven't submitted a request in the last two weeks?", "Which clients gave the lowest review scores this month?"

Collections: "Which unpaid invoices are more than 30 days old?"

Tag reporting: "How many requests did the Website tag get this month?", "Show me all requests tagged Urgent that are in progress."

Client recaps: "Summarize everything that moved on Northwind Studio's account this month and post it as a comment on their latest request."

Custom reports: "Create a chart of requests submitted this year by month.", "How many requests did each team member complete last month?"

Bulk actions: "Reassign all unassigned requests to Sam.", "Send a payment reminder to every client with a pending invoice."

What the MCP server cannot do

For these, go to the portal:

  • Change the client, service, billing, credits or time estimates of a request.

  • Change custom field values on an existing request.

  • Move a client to another organization, or delete or restore a client.

  • Refund, edit or delete an invoice, or apply a coupon.

  • Cancel or change a running subscription. Cancelling settles payments, so it is done in the portal.

  • Upload files. Attachments are uploaded in the portal.

  • Edit or delete a comment, or schedule a comment for later.

  • Rate a request or leave a tip. Reviews are left in the portal.

Before you start

  • You don't need IDs: say "Hannah from Northwind Studio" and the AI looks up the right client.

  • Some actions can't be undone: posting a comment, marking a request as completed, sending an invoice, marking it paid, sending a payment reminder and creating a client with an invitation email all reach people right away. Claude and ChatGPT usually ask you to confirm before the AI posts a comment or adds a subscription. Ask for exactly what you want.

  • Comments are client-facing by default: say "internal note" when only your team should see it.

  • Invoices start as drafts: the AI only issues an invoice to the client when you ask it to send or issue it.

  • New clients get an invitation by default: say "don't send the invitation" to open the account quietly.

  • Tracked time follows time approval: if time entry approval is on, pending and rejected time is not included in tracked hours.

  • The AI sees what you can see: it works as your user, so anything outside your access stays hidden from it too.

How to see and revoke connections

Everyone who can connect sees their own connections on the AI assistants tab of their profile, under Your connections, with when they authorized each one and when it expires. On the MCP server tab, Authorized connections lists the AI apps you approved yourself.

Portal admins and admins also see Every connection in this portal on their AI assistants tab: each connection with the person's name, their role and the AI app. Use it to cut off a team member who left, or a client you no longer work with.

  1. Click the three dots next to your name at the bottom of the sidebar, then click Your profile.

  2. Open the AI assistants tab.

  3. Click Revoke next to a connection, then Confirm. You see AI assistant connection revoked.

    AI assistants tab for a portal admin listing your connections and every connection in this portal

The AI app loses access right away and can't reconnect on its own. To use it again, the person connects again from their AI app.

FAQ

A team member or client doesn't see the AI assistants tab. Why?

Check that:

  • Their role is allowed under Who can connect on the MCP server tab.

  • Your portal is on the Pro or Core plan.

The tab is also hidden while you impersonate someone, so you can't connect an AI app on another person's behalf. They connect it themselves.

Does the AI app show ManyRequests to my clients?

Not with your portal's own link. That link uses your portal address, the approval screen shows your portal logo, and the MCP server introduces itself to the AI app with your portal name.

With the ManyRequests link (https://app.manyrequests.com/mcp) or the connector directory, yes: the AI app lists the connector as ManyRequests, and the page where the person types your portal address shows the ManyRequests logo. The login and approval screens after it are your portal's.

If I turn off a role while impersonating someone, do I lose my own connection?

No. Only the connections of the people whose role is no longer allowed are cut. Portal admins and admins always keep theirs.

Is the MCP server the same as an API token?

No. An API token doesn't work for the MCP server, and the other way round. See How to get started with the ManyRequests API.

Prefer automations over chat? See How to connect Zapier to your ManyRequests account.

Did this answer your question?