MCP client
ChatGPT — a custom connector, and the tool limits to expect
Connect Attensira to ChatGPT as a custom connector over OAuth, and understand why some write tools may be unavailable for reasons that are not scope.
Karl-Gustav Kallasmaa, Founder & CEOLast updated What you configure
The exact settings, as our documentation publishes them.
Settings → Connectors → Create
Type: custom / developer-mode connector
Server URL: https://mcp.attensira.com/mcp
Auth: OAuth
Connecting it
In order, with the reason each step exists rather than only the instruction.
- 1
Create the connector
Under Settings, then Connectors, then Create, choose a custom or developer-mode connector and enter the server URL.
- 2
Select OAuth and fill in nothing
At the authentication step select OAuth. Attensira publishes standard OAuth discovery, so there is no client ID and no secret for you to supply. Save and complete the sign-in in the browser.
- 3
Pick the workspace
The consent screen names the application, the workspace and the access it wants. A connection reaches one workspace and cannot be moved afterwards, so choose deliberately rather than accepting the first option.
- 4
Enable it per conversation
Enable the connector from the tools menu in each conversation you want it in. Then ask for get_account as a smoke test — a reply naming your workspace confirms the credential is bound.
What it lets you do
- Reads the workspace's tracking data inside ChatGPT — mention and citation rates, which prompts are losing, which domains models cite in your category.
- Uses standard OAuth discovery, so connecting involves no client registration and no secret to store anywhere.
- Exposes the same eighteen tools as every other client, subject to whatever the ChatGPT surface itself permits.
What it cannot do
An integration page that lists only capabilities is an advertisement. This is the half you are evaluating.
- Custom connector availability in ChatGPT depends on your plan and workspace settings, and some surfaces restrict which tools a connector may call at all. Neither is something Attensira can change from its side.
- A write tool being unavailable in ChatGPT is not necessarily a scope problem. Check what you actually granted under connected applications before assuming the connection is wrong, because the restriction may be imposed by the client.
- A connection reaches exactly one workspace and cannot be retargeted. Reaching a second one means creating a second connector and approving it separately.
ChatGPT connects to Attensira as a custom connector over OAuth. The mechanics are short; the part worth reading carefully is what happens when a tool you expected to be there is not, because the usual explanation is not the one people reach for.
Creating it
Under Settings, then Connectors, then Create, choose the custom or developer-mode connector and enter the server URL. At the authentication step, select OAuth and supply nothing else. Attensira publishes standard OAuth discovery, so there is no client ID and no secret for you to find or paste.
Save, complete the sign-in, choose the workspace at the consent screen, and enable the connector from the tools menu in the conversations you want it in.
Two different reasons a tool can be missing
This is the section that matters, and it separates two failures that look identical from inside a chat.
The first is scope. Read access covers metrics, prompts, pages, sessions, automations and account details. Everything that changes the workspace or spends credits needs write, and if you approved read only, a write tool will be refused — a valid credential being told no. That is fixed by revoking the connection and re-approving it with wider access, because scope cannot be widened in place.
The second is the client. Some ChatGPT surfaces restrict which tools a connector may call at all, independently of what you granted. When a write tool like adding prompts is unavailable, that restriction is the likelier explanation, and it is not something any amount of reconnecting on the Attensira side will change.
The order to check them in is the cheap one first: look at what you actually granted under connected applications before assuming the connection is wrong. If it says read and write and the tool is still absent, the restriction is on the ChatGPT side.
Availability is a plan question
Whether you can create a custom connector at all depends on your ChatGPT plan and workspace settings. That is worth establishing before debugging anything, because "the option does not exist" and "the option exists and the connector does not work" have nothing in common.
What it is good at
The natural fit here is the analytical conversation rather than the operational one. Pulling mention and citation rates for a window, comparing them against the previous one, asking which sources the models lean on in your category, working out which prompts you lose and to whom — all of that is read-only and lands squarely inside what a chat client does well.
Because the credential names the workspace, no request carries a workspace id, and there is nothing in a conversation that can be steered towards somebody else's data.
Revoking, and what revoking does not do
Connections you have approved are listed under Settings, then Connections, and pressing Revoke stops the client's next refresh. It does not instantly invalidate an access token the client is already holding — that stays valid until it expires, which is at most an hour.
The practical reading of that: revoking is the right first move when a credential has leaked, and it should be followed by assuming up to an hour of remaining access rather than none. For a laptop you no longer use or a client you were merely trying out, an hour is irrelevant and revoking first and asking later is the cheap direction.
A note on where the boundary sits
It is worth internalising which side of this integration owns which decision. Attensira owns the endpoint, the toolset, the scopes and the workspace binding. ChatGPT owns whether custom connectors exist for your plan, whether an administrator has permitted them, and which tools a given surface will let a connector call. When something does not work, deciding which of those two lists the problem belongs to is the first and cheapest diagnostic step, and it usually resolves the question without any configuration being changed at all.
Questions people ask
- Do I need a client ID or secret?
- No. Attensira publishes standard OAuth discovery, so selecting OAuth is enough and there is nothing to fill in.
- Write tools like add_prompts are missing. Is my scope wrong?
- Not necessarily. Some ChatGPT surfaces restrict which tools a connector may call, and that is a client-side restriction rather than a scope problem. Check what you granted under connected applications before changing anything.
- Why can I not create a custom connector at all?
- Availability depends on your ChatGPT plan and workspace settings. That is decided on the ChatGPT side.
- Can I use a static API key here?
- The documented path for ChatGPT is OAuth. Static keys exist for headless callers that cannot complete a browser sign-in, which is not the situation in a chat client.
- Where do I see and revoke what I granted?
- Under Settings then Connections in Attensira. Revoking stops the next refresh; a token the client already holds remains valid for at most an hour.
Sources
Every factual statement above, with the page it came from and the date that page was read.
Attensira's connect documentation states that custom MCP connectors in ChatGPT depend on plan and workspace settings, that some surfaces restrict which tools a connector may call, and that an unavailable write tool is a ChatGPT-side restriction rather than a scope problem.
docs.attensira.com · retrieved
“If write tools such as add_prompts are unavailable, that is a ChatGPT-side restriction, not a scope problem”
Attensira's connect documentation states that Attensira publishes standard OAuth discovery, so there is no client ID or secret for a ChatGPT connector to fill in.
docs.attensira.com · retrieved
“Attensira publishes standard OAuth discovery, so there is no client ID or secret for you to fill in.”
Attensira's connections documentation states that revoking stops refreshes immediately but an access token already held stays valid until it expires, which is at most 60 minutes.
docs.attensira.com · retrieved
“It stops refreshes immediately, but an access token the client is already holding stays valid until it expires, which is at most 60 minutes.”