Cadient Talent’s MCP
Cadient Talent’s Model Context Protocol (MCP) lets approved AI assistants connect directly to Cadient Talent, so your team can work with live hiring data inside the tools they already use. Once connected, an assistant can surface insights, run analyses and answer questions using your Cadient Talent data as context. There’s no need to switch tabs or build custom reports.
MCP video overview
What is Cadient Talent’s MCP?
Cadient Talent’s MCP gives approved AI assistants a secure, structured way to work with your hiring data. The MCP acts as a layer on top of the Cadient Talent API, translating natural language requests into structured API calls, scoped to your existing permissions. This makes it possible to retrieve data and answer questions without navigating the grids directly or building a custom integration.
MCP stands for Model Context Protocol, an open standard AI assistants use to reach outside tools and data sources. Nobody on your team needs to understand it to use the agent; it is the plumbing, not the product.
Why use Cadient Talent’s MCP?
Cadient Talent’s MCP gives teams faster access to live hiring data inside the tools they already use. Because it is built and maintained by Cadient, access is governed by your organisation’s existing Cadient Talent permissions rather than by a new set of rules.
With Cadient Talent’s MCP, teams can:
- Use live Cadient Talent data inside approved AI assistants, with no manual exports or custom integrations
- Keep access aligned with existing Cadient Talent permissions, location visibility and organisational governance
- Give field and hiring managers self-serve answers without training them on a reporting tool
- Support auditability with Cadient-managed access controls and logging for all MCP activity
- Avoid standing up a second copy of hiring data in a warehouse or BI layer just to answer questions
MCP use cases
Cadient Talent’s MCP is useful for recruiting operations, pipeline visibility and hiring analysis. Representative examples include:
- Building a live pipeline view for hiring managers across active requisitions
- Seeing how long candidates have sat in each status and flagging the ones that have stalled
- Identifying which locations or requisitions are the bottleneck across an open pipeline
- Finding requisitions that have been open past a threshold with nobody in process
- Auditing requisitions that have no live posting attached
- Comparing intake and conversion by source to see which ones actually produce hires
- Checking where a specific person’s onboarding has stopped
- Reconciling a reported number against the filters that produced it
A shorter version of the same list, grouped by who tends to ask:
Recruiters
- Everyone in process for a requisition, and how long each has been there
- Candidates stalled at one status past a threshold
- Last week’s intake for a set of locations
Hiring & field managers
- Their own open requisitions with days-open and count in process
- Which posting is actually driving applicants at their location
- A requisition with zero candidates in process, before it becomes a staffing problem
Talent acquisition leadership
- Average time-to-fill by quarter, split hourly vs. salaried
- Funnel counts by stage, and where the biggest drop-off sits
- Hires by location or source over a named window
Recruiting operations
- Locations where intake dropped after a posting change
- Month-over-month intake comparisons by region
- Data-quality gaps that distort reporting
Before you begin
Before connecting Cadient Talent’s MCP, confirm the following:
- You have an active Cadient Talent account: the agent uses the same credentials, and deactivated accounts have no access
- You know your customer ID (your organisation’s Cadient identifier). If you don’t, ask Cadient customer support at support@cadienttalent.com. It identifies your company, not you
- Your AI assistant supports MCP over Streamable HTTP transport
- For the first connection at your company, Cadient support may need to register the assistant’s redirect address once. Email support@cadienttalent.com if the sign-in flow does not complete. See Troubleshooting
Supported AI assistants
Cadient Talent’s MCP currently supports:
- Claude: web and desktop
- Claude Code (CLI)
- ChatGPT: Business, Enterprise and Edu, and Pro with read and fetch permissions. Web only, and a workspace admin has to enable developer mode first
- Any other MCP-compatible assistant that speaks Streamable HTTP transport
To connect an assistant not listed above, email its redirect address to Cadient support at support@cadienttalent.com so it can be added to the allowlist. That step is required for the sign-in flow to complete successfully.
Setup steps for both are below. Switch between Claude and ChatGPT with the tabs.
Connect Cadient Talent’s MCP
Connecting takes about two minutes and is done once per person, per assistant. Nothing is installed and no new account is created. The server address is the same for every assistant. Only the screen you paste it into changes. Both assistants reach the same 35 read-only capabilities, and only the setup path differs.
Server addresses
Enter the address exactly as shown, including https:// and the /mcp path, with no trailing slash. Use production unless you have been asked to test.
| Environment | Connector name | Server address |
|---|---|---|
| Production | Cadient | https://mcp.cadienttalent.com/mcp |
| UAT / testing | Cadient UAT | https://mcpuat.cadienttalent.com/mcp |
Pick your assistant
Any individual user can add the connector themselves: web, desktop or the Claude Code CLI. No admin step, about two minutes.
Step 1: Open the Connectors screen
In Claude, open Settings → Connectors. You’ll see the connectors already available to you with their type and connection status. Select Add at the top right, then Add custom connector. Discover, next to it, lists pre-built connectors. Cadient Talent’s MCP is added as a custom connector.
Step 2: Enter the name and address
In the Add custom connector dialog, give the connector a name (Cadient for production, or Cadient UAT for the test environment), and paste the matching server address. Then select Continue.
Step 3: Sign in with your Cadient Talent credentials
Back on the Connectors screen, select Connect against the new entry. Claude redirects you to the Cadient sign-in page. Enter three details, then submit and approve the connection when your assistant asks.
| Field | What to enter |
|---|---|
| Customer ID | Your organisation’s Cadient customer ID. Don’t know it? Ask Cadient customer support at support@cadienttalent.com. It identifies your company, not you. |
| Username | Your Cadient Talent username, the one you use to log in today. |
| Password | Your Cadient Talent password. It is checked against Cadient Talent and never stored. |
In the Claude Code CLI
claude mcp add --transport http cadient https://mcp.cadienttalent.com/mcp
Then follow the sign-in prompt. Any other MCP-compatible assistant works the same way: give it the server address and it discovers the sign-in flow automatically.
A workspace admin turns on developer mode once and adds the plugin. Available on ChatGPT Business, Enterprise and Edu, and on Pro with read and fetch permissions. Web only.
Step 1: Turn on developer mode
Open the account menu at the foot of the ChatGPT sidebar and select Settings.
Go to Security and login and switch Developer mode on. This is a workspace-level setting, so it is done once by an admin rather than by every user.
Step 2: Add the plugin
Go to Plugins and select + at the top right to open the New Plugin form.
Fill the form in as follows, tick I understand and want to continue, then select Create.
| Field | What to enter |
|---|---|
| Name | Cadient talent |
| Description | Cadient talent’s MCP |
| Server URL | The address from the table above, including /mcp |
| Authentication | OAuth |
Step 3: Sign in with your Cadient Talent credentials
ChatGPT shows an Add Cadient talent to ChatGPT confirmation. Select Sign in with Cadient talent.
Enter your customer ID, username and password, the same ones you use in Cadient Talent today. Don’t know your customer ID? Ask Cadient customer support at support@cadienttalent.com. It identifies your company, not you.
Approve the three permissions ChatGPT asks for: your user profile, your email address, and offline access, which is what keeps the connection live between sessions.
The plugin now shows as connected. Permissions defaults to Always ask, so ChatGPT checks with you before each use.
First things to try
- “What can you tell me about my hiring data?” (the agent describes what it can do)
- “Who am I in Cadient Talent?” (confirms it is running as you, with your visibility)
- “List my open requisitions.”
- “How many applicants did we get in the last 7 days?”
How access is governed
Three things decide what any given person can get an answer to:
- Their own Cadient Talent permissions. Location visibility, role-based access and “my reqs vs. all reqs” scoping all carry across unchanged. The agent cannot widen them.
- The supported capability set. There are 35 read-only capabilities today. Anything outside that set is not reachable, regardless of permissions.
- Sensitive-field suppression. Government IDs, dates of birth, tax data (W4 / I9), EEO data and identity documents are blocked at two independent layers and are never returned to any user.
There is no separate scope console to configure and no per-user provisioning step. Every capability is read-only, so there is no write permission for an admin to grant or withhold. Access simply follows Cadient Talent. Add a person there and the agent is available to them; deactivate them there and their access ends the same minute.
Two defaults to know
The agent deliberately mirrors the product’s grids rather than the raw database. Knowing these two defaults is the difference between trusting a number and re-checking it.
Hourly or salaried
Application questions default to hourly, matching the Applicant grid. Candidate questions default to salaried, matching the Candidate grid. Ask explicitly (“including salaried”, “hourly and salaried both”) for the other population or for everything.
Your requisitions or all requisitions
Requisition-scoped questions default to your requisitions, the same way the product scopes “My Requisitions”. Ask for “all requisitions” to widen. If your role does not permit company-wide visibility, the agent tells you it answered from your own scope rather than presenting it as the company-wide number.
“Not found” and “not visible to you” look the same
This is deliberate. If a record you expect is reported as not found, check it in Cadient Talent. If you cannot see it there either, it is outside your visibility.
Get better results
Cadient Talent’s MCP uses live data. The cleaner and more structured that data is, the better the answers. If you are not sure where to start, begin with statuses and postings, because those two areas have the biggest effect on output quality.
Keep statuses moving
Move people through statuses promptly and close out those who are no longer active. Stale statuses are the single most common reason a pipeline answer looks wrong.
Keep postings tied to requisitions
A requisition with no live posting attached will read as having no intake. Auditing for that is itself a good first question to ask the agent.
Use consistent location and role naming
Consistent naming across locations lets the agent group and compare accurately. Inconsistent naming shows up as suspiciously small counts.
Prefer structured fields over free text
Single-select, multi-select, yes/no, number, date and currency fields are far easier to analyse than free-text notes.
Say what you mean by the window
“This month”, “last 30 days” and “this quarter” are three different numbers. Naming the window in the question is the fastest way to make an answer reconcilable against a grid.
Review permissions deliberately
The agent respects existing permissions exactly. Thoughtful permission setup is what lets people get the answers they need without widening access to hiring data.
Security & data handling
| Control | How it works |
|---|---|
| Your data doesn’t move | Every answer is read at question time. There is no export, no cached copy, and no separate analytics or data-warehouse store of your hiring data. |
| Read-only by design | All 35 capabilities read data. None create, edit or delete a record. There is no write path. |
| Identity stays in Cadient Talent | Sign-in verifies credentials against Cadient Talent and stores nothing: no user record, no password, nothing to de-provision when someone leaves. |
| Every request runs as the person asking | Each call carries a short-lived (about two minutes), cryptographically signed statement of who is asking. No shared service account, and no way to ask “as” someone else. |
| Company isolation | Enforced by verified credentials, never by a URL or header a caller could spoof. One platform serves many organisations; each sees only its own data. |
| Sensitive fields suppressed | Government IDs, dates of birth, tax data (W4 / I9), EEO data and identity documents are blocked at two independent layers. |
| Accounts can’t be locked from here | The sign-in page allows fewer consecutive wrong-password attempts than Cadient Talent does, so repeated failures block the sign-in page for 15 minutes, never your Cadient Talent account. |
| Everything is audited | Every question is logged with who asked, which capability ran, and which organisation it belonged to. |
Troubleshooting
I don’t have access to Cadient Talent’s MCP
If you cannot connect, confirm the following:
- Your Cadient Talent account is active; deactivated accounts have no access
- You have your organisation’s customer ID
- Your AI assistant supports MCP over Streamable HTTP transport
- The redirect address for your assistant has been allowlisted by Cadient, if required
Sign-in says the customer ID, username or password is wrong
One of the three didn’t check out. Re-check all three: the customer ID is your organisation’s identifier, not your username. Ask Cadient customer support at support@cadienttalent.com if you are unsure of it.
“Too many sign-in attempts”
A protective limit tripped after repeated failures. It trips before Cadient Talent’s own lockout, so your account itself is unaffected and you can still log into the product. Wait 15 minutes and try again with the correct password.
Sign-in says “service unavailable” rather than wrong password
This points to a service fault rather than anything wrong with your credentials. Try again shortly. If it persists, email support@cadienttalent.com with the time it happened.
I signed in with the wrong account or environment
Disconnect the Cadient connector in your assistant, sign out of Cadient Talent, and start the connection again with the right credentials. If you are switching between UAT and production, sign out of the current environment before reconnecting.
It was working, now it asks me to sign in again
The session expired, which is normal. Sign in again with the same three fields.
“Invalid parameter: redirect_uri”
Your assistant’s callback address has not been registered yet. Email support@cadienttalent.com with the exact error and the name of your assistant, and support can add it.
My assistant is missing capabilities I expected
There are 35 read-only capabilities today. If something you expected is missing, check whether it is a write action (not supported in this release), whether your Cadient Talent permissions allow it, and whether the record is inside your visibility. Supported capabilities will continue to expand over time.
A record I know exists comes back as “not found”
Either it does not exist, or it is outside your visibility, and the two are deliberately indistinguishable. Check the record in Cadient Talent; if you cannot see it there either, that is your answer.
A count doesn’t match the grid I’m looking at
Most often the hourly / salaried default, a different date window, or “mine vs. all” scoping. Ask the agent to restate the filters it used, then align them with the grid’s. See Two defaults to know.
Someone who left the company can no longer connect
Working as designed. Access follows Cadient Talent: deactivated there means no access here, immediately.
Frequently asked questions
Who can use Cadient Talent’s MCP?
Any active Cadient Talent user. There is no per-user provisioning, no invite and no account creation. People add the connector and sign in with the credentials they already have.
Is it generally available?
Yes. It is documented and available to Cadient Talent customers today. Your account team can help with rollout for your organisation.
Can it take actions, or is it read-only?
It is read-only. All 35 capabilities read data; none create, edit or delete a record. There is no write path.
Does access follow our existing permissions?
Yes. Every answer is filtered through the asking person’s own Cadient Talent permissions: the same location visibility and the same “my reqs vs. all reqs” scoping. Nobody can reach data beyond what their permissions already allow.
Which AI assistants are supported?
Claude on web and desktop, the Claude Code CLI, ChatGPT on the web, and any MCP-compatible assistant that speaks Streamable HTTP transport. See Supported AI assistants.
Can we use ChatGPT instead of Claude?
Yes. Cadient Talent’s MCP works with any MCP-compatible assistant that speaks Streamable HTTP transport, and that includes ChatGPT. The setup differs: a ChatGPT workspace admin has to enable developer mode and publish the connector before members can connect. See the ChatGPT setup steps.
Where are our credentials stored?
Nowhere. The password is checked against Cadient Talent at sign-in and immediately forgotten. There is no second user database and no synced password.
Is any of our hiring data copied or cached?
No. Every answer is read from Cadient Talent at the moment the question is asked. No export, no cached copy, and no separate analytics or data-warehouse store.
Can it return SSNs, dates of birth or EEO data?
No. Those fields, along with tax data (W4 / I9) and identity documents, are blocked at two independent layers and are never returned by any capability.
Is there an admin screen for managing scopes?
No, and there is nothing it would control: every capability is read-only and bound to the user’s existing permissions, so there is no write access for an admin to grant or withhold. If action-taking is added in a future release, admin controls will come with it.
Do we have to onboard separately to use it?
No. If your organisation uses Cadient Talent, this is already available to you. There is no separate registration step and nothing to configure.
Is there an audit trail?
Yes. Every question is logged with who asked, which capability ran, and which organisation it belonged to.
What’s the difference between the production and UAT addresses?
Production reads your live data; UAT is for testing. Each has its own connector entry, so a person can have both connected at once without confusion. Use production unless you have been asked to test.
Does using this cost extra?
There is no separate provisioning step to pay for. Commercial terms are handled through your existing Cadient Talent agreement, and your account team can confirm what applies to you.