Quick answer: Bullhorn does not offer an official MCP server, so there is no first party way to connect Claude or ChatGPT to your Bullhorn database. Agencies that want it use a third party bridge: StackOne, Knit, CData, Ribbon or Osher’s free open source server. Every one of them needs Bullhorn REST API credentials, which you request from Bullhorn Support.
Best fit by need: Osher if you want free and read only, CData if you want a simple monthly price, StackOne or Knit if you have someone technical, Ribbon if you already screen candidates with AI voice interviews.
Who this is for: agency owners and ops leads running Bullhorn who want AI working on their own data.
Does Bullhorn have an official MCP server?
No. Bullhorn puts its AI inside its own product under the Amplify name. At Engage Boston on May 28, 2026 it launched Amplify Digital Workers and Amplify Chat, a chat interface over live Bullhorn data for Amplify customers, according to Carly’s Bullhorn breakdown. That is useful if you pay for Amplify, but it is not an MCP server you can plug into Claude, ChatGPT or Copilot.
So the question for a Bullhorn agency is not whether to use MCP, but which bridge to trust with your database. We cover the bigger picture in agencies and MCP access.
What do you need before any Bullhorn MCP option works?
REST API credentials. Bullhorn does not hand these out through a settings screen. Its own knowledge base says the credentials are not self service and a request must go to Bullhorn Support, which then delivers a Client ID, Client Secret and API Username, per Bullhorn’s credential guide. Each environment, staging or production, needs its own pair.
Bullhorn publishes no price for API access, and whether it costs anything depends on your agreement with Bullhorn, as JobsPipe notes. Ask your account manager before you budget.
Which Bullhorn user should the AI connect as?
This decision matters more than which vendor you pick. The bridge sees whatever its Bullhorn user can see. Create a dedicated API user with the narrowest access the job needs, never an admin account. Our guide to protecting your database walks through the permission settings.
How do the Bullhorn MCP options compare?
| Option | Type | Read or write | Starting price | Best for |
|---|---|---|---|---|
| Osher | Free, open source, you host it | Read only, six tools | Free | Testing with zero risk of edits |
| CData Connect AI | Hosted MCP platform | Live queries, plus write operations you approve | $99 a month, or $79 billed annually | A simple monthly bill, no code |
| StackOne | Hosted, managed | Read and write, 80 prebuilt actions | Free starter, Team $600 a month | Teams that want managed auth and audit |
| Knit | Hosted | Read and write | Free Launchpad, Individual $29 a month | A cheap technical trial |
| Ribbon | Hosted, tied to AI interviews | Reads jobs and candidates; writes interview results and notes | Plans from $199 a month | Agencies already screening with Ribbon |
Prices were checked on each vendor’s own page in November 2026.
Osher: free and read only
Osher Digital’s server is free and open source, runs in Python on your own machine, and keeps your credentials in a local file. It only reads: the page says it does no create, update or delete, and it ships six tools such as list jobs, list candidates and search entities, per Osher’s release post. It is the lowest risk way to see what Claude can do with your data. You will need someone comfortable installing Python.
CData Connect AI: the simplest monthly bill
CData sells a hosted MCP platform with a published price: Standard is $99 a month or $79 billed annually for one user and one standard data source, and Growth is $199 a month or $159 billed annually, per CData’s pricing page. CData says your existing Bullhorn permissions follow the data into the AI, with full query logging, and that you can expose only the write operations you approve, per CData’s Bullhorn page. It is listed in Claude’s connector directory and offers a 30 day trial. Confirm which pricing tier Bullhorn falls into before you buy.
StackOne: managed actions with security controls
StackOne runs a hosted Bullhorn MCP server with 80 prebuilt actions and handles authentication, tool execution and security for you, per StackOne’s Bullhorn page. Its Gateway Starter plan is free with 1,000 credits per seat a month, and Gateway Team is $600 a month with 5,000 credits per seat, per StackOne pricing. The page does not say whether that $600 is per seat or flat, so ask.
Knit: the cheapest hosted trial
Knit runs the Bullhorn MCP server for you with no server to deploy. Its MCP Servers product has a free Launchpad tier with 100 API calls a month and an Individual tier at $29 a month with 5,000 calls and unlimited servers, per Knit pricing. Its Bullhorn page lists a handful of actions and says Knit can build missing calls on request.
Ribbon: MCP that comes with AI screening
Ribbon’s MCP gives an agent live access to job orders, submissions, candidates, client contacts, appointments and notes, scoped to your Bullhorn permissions. Writes, such as moving a submission or attaching an interview result, go through a separate audited path, per Ribbon’s MCP post. Ribbon’s plans start at $199 a month, with ATS integration on higher tiers per Ribbon pricing, so it makes most sense if you want AI voice screening anyway.
Which Bullhorn MCP option should you pick?
| If you want to | Pick | Why |
|---|---|---|
| Try it this week with no risk of edits | Osher | Free and read only |
| Give one ops person a no code setup | CData Connect AI | Published monthly price, hosted |
| Roll out to a team with audit and controls | StackOne | Managed auth, per action permissions |
| Prototype cheaply with a developer | Knit | Free tier, then $29 a month |
| Screen candidates and log results into Bullhorn | Ribbon | Interview results written back automatically |
Whatever you pick, start read only. Let the team ask questions of the data for a few weeks before you let any AI write to Bullhorn.
How do you set up a Bullhorn MCP connection?
The steps are the same whichever bridge you choose:
- Request API credentials. Open a ticket with Bullhorn Support for a Client ID, Client Secret and API Username. Ask for staging first if you have it.
- Create a narrow API user. Give it only the record types and fields the AI needs. Leave out pay rates, notes marked private and anything you would not want summarized in a chat window.
- Connect the bridge. Enter the credentials in the vendor’s dashboard, or in a local settings file for Osher. Hosted options usually also need Bullhorn to register their redirect address on your key.
- Add it to your AI assistant. In Claude, ChatGPT or Copilot, add the bridge as a custom connector using the URL the vendor gives you, then sign in.
- Test read only questions. Ask for open job orders, recent submissions or a candidate summary, and compare the answers with Bullhorn before anyone relies on them.
Budget a few days for step one. Bullhorn handles credential requests through its support queue, and each environment is a separate ticket.
What are the risks of connecting Bullhorn to AI?
Three come up again and again:
- Speed of extraction. An AI with database access can pull far more records in one request than a recruiter could copy by hand. Lock down the API user’s permissions first. See database protection steps.
- Hidden instructions in resumes. When an AI reads candidate files, text planted in a resume can try to steer it. Read our guide to resume prompt injection.
- Where the data goes. Check whether your AI plan trains on your data and what your candidate privacy notice allows. Our candidate data guide covers the rules.
Put the rules in writing before anyone connects. Our agency AI policy template has a section for MCP connections you can copy.
Quick answers
Is there a Bullhorn MCP server?
Not from Bullhorn. Third parties such as StackOne, Knit, CData, Ribbon and Osher offer MCP access to Bullhorn data.
How do I get Bullhorn API credentials?
Open a ticket with Bullhorn Support. They issue a Client ID, Client Secret and API Username, one set per environment.
What is the cheapest way to connect Claude to Bullhorn?
Osher’s open source server is free and read only. Knit’s hosted Launchpad tier is also free, with 100 API calls a month.
Can AI update records in Bullhorn?
Yes, through StackOne, Knit, CData or Ribbon. Start read only and add write access later, one action at a time.
Is Amplify Chat the same as MCP?
No. Amplify Chat works inside Bullhorn for Amplify customers. MCP lets an outside assistant such as Claude or ChatGPT reach the data.
Which AI assistant should we use with Bullhorn?
Any MCP client works. Compare plans and admin controls in our assistant comparison.
The bottom line
Bullhorn agencies can connect AI to their database today, but only through a third party, and every route starts with an API credential request to Bullhorn Support. Start with a read only option, connect through a narrow API user, and write your rules down first. For every guide in this series, see MCP for staffing agencies.
Sources: Carly, Bullhorn KB, JobsPipe, Osher, CData pricing, CData Bullhorn, StackOne Bullhorn, StackOne pricing, Knit Bullhorn, Knit pricing, Ribbon MCP, Ribbon pricing
Free newsletter
Get the Work Insiders weekly
One email a week: the recruiting and HR tech news that mattered, the tools we just added to the directory, and one practical guide. No spam, and you can unsubscribe in one click.