Sales ops · Production

HubSpot launches the first CRM Deep Research Connector with ChatGPT via a remote MCP server

The problem

HubSpot's previous local MCP server required manual token creation and client configuration, limiting CRM AI access to technical developers. In parallel, HubSpot's public APIs were not originally designed to enforce user-level permissions, making a secure user-scoped AI connector difficult to deliver at scale across thousands of APIs maintained by hundreds of teams.

First attempt

The prior local MCP server approach, published on NPM, required users to manually create Private App Access Tokens and configure MCP clients, introducing friction that blocked non-technical users and limited adoption.

Workflow diagram · grounded in source
1
User authenticates via OAuth
trigger
“they can access capabilities through their browser using trusted and reliable authentication flows like OAuth”
2
MCP tools auto-discovered
integration
“our MCP server periodically polls for new or updated services deployed with these annotations. The system then fetches their specifications and automatically converts each RPC to a MCP spec-compliant tool. Finally, these wrapped RPCs are…”
3
User permissions enforced
validation
“Extract the relevant scopes of the access_token ... Fetch user-specific scopes for the user ... Replay those permissions to internal APIs where user permissions are respected for data within the CRM”
4
AI generates DSL search query
ai_action
“Each search query consists of space-separated tokens following a key[:operator]:value pattern... What makes this work well with AI models is its predictable structure and clear examples”
5
CRM research results returned
output
“we support contacts, companies, tickets, and deals in our Deep Research Connector”
Reported outcome

HubSpot shipped a browser-accessible remote MCP server that democratizes AI-powered CRM access for more than 250,000 businesses with enterprise-grade security and user-level permissions enforced, delivered as a production-ready feature in a matter of weeks.

Reported metrics
businesses with democratized AI CRM accessmore than 250,000
Implementation time to productiona matter of weeks
API latency overhead for runtime permission computationa few dozen milliseconds slower
Reported stack
ChatGPTModel Context ProtocolJava MCP SDKDropwizardOAuthCHIRPCloudflare WorkersOpenAI
Source
https://product.hubspot.com/blog/unlocking-deep-research-crm-connector-for-chatgpt
Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

HubSpot shipped a browser-accessible remote MCP server that democratizes AI-powered CRM access for more than 250,000 businesses with enterprise-grade security and user-level permissions enforced, delivered as a produc…

What tools did this team use?

ChatGPT, Model Context Protocol, Java MCP SDK, Dropwizard, OAuth, CHIRP, Cloudflare Workers, OpenAI.

What results were reported?

businesses with democratized AI CRM access: more than 250,000; Implementation time to production: a matter of weeks; API latency overhead for runtime permission computation: a few dozen milliseconds slower (source-reported, not independently verified).

What failed first in this deployment?

The prior local MCP server approach, published on NPM, required users to manually create Private App Access Tokens and configure MCP clients, introducing friction that blocked non-technical users and limited adoption.

How is this sales ops AI workflow structured?

User authenticates via OAuth → MCP tools auto-discovered → User permissions enforced → AI generates DSL search query → CRM research results returned.

Related sales ops cases