The Claude Connector within Dobbin:
🔸 Overview
• Connects Dobbin to Claude via the Model Context Protocol (MCP) to securely push your proprietary company memory directly into Claude's interface.
🔸 Value
• Eliminates the need to constantly re-educate generic AI tools by supplying Dobbin's proprietary judgment and memory while Claude handles the drafting and execution.
🔸 Setup
1. Have your Claude admin open Claude and navigate to Customize (the briefcase icon) or go directly to claude.ai/customize/connectors.
2. Click Connectors, click the Add connector button, name the connector "Dobbin," and paste your specific scoped URL (generated in Dobbin HQ under Control Center > Connectors).
3. Save the connector and complete the sign-in flow to authenticate.
4. As the tool is being used, select "Allow" for the permissioning.
🔸 Security
• Acts as a zero-trust orchestration layer that provides governed, controlled disclosure—Claude receives only the isolated snippets needed to answer the prompt, ensuring you never surrender your entire corpus to a public model.
• Enforces your existing permissions, backed by SOC 2 Type 2 and ISO 27001 certified infrastructure.
• Passes all requests through a secure Cloudflare API gateway that strips end-user cookies and injects cryptographic credentials.
• Ensures customer data is never used to train public models.
🔸 Technical Safeguards & Edge Cases
• Clarify to users that working in Claude is a one-way recall—Dobbin provides its full knowledge base but does not learn from or ingest any Claude conversations.
• Positions Dobbin not as a feature-for-feature competitor to Claude, but as the governed delivery mechanism that renders Claude safe and on-brand enough to use in an enterprise at all.
🔸 Good Prompts
• "Use Dobbin to pull our objection-handling framework, then draft a reply to this client pushing back on our timeline."
• "Run this pitch past Dobbin and tell me the blind spots we'd flag internally before it goes out."
• "Based on what Dobbin knows about our customers, which of these two campaign directions fits us better?"
🔸 Bad Prompts
• "Help me with this client email."
Why it fails: No tool is named, so Claude may never call Dobbin at all, resulting in generic, ungrounded output while the strategic layer sits idle.
• "Export our entire brand guidelines."
Why it fails: The zero-trust layer is designed to hand Claude only the isolated snippet the immediate prompt needs; bulk extraction fights the architecture, so you should ask a scoped question and let it return the relevant slice instead.
• "Use Dobbin to write me a headline."
Why it fails: Nothing here touches company judgment or memory, adding approval friction for zero strategic payoff and often calling the wrong tool for generic work that Claude can do alone.
• "Update the deal in our CRM."
Why it fails: Neither Dobbin nor Claude can take actions in outside systems, meaning they cannot modify external databases or CRMs.
🔸 FAQs
• Q: Why not just paste my strategy into Claude myself?
A: Pasting strategy directly hands a general-purpose tool your entire brain and bypasses all governance. The connector allows you to point a commodity model at proprietary strategy through controlled disclosure, answering the question without losing control of who sees what.
• Q: Do I need to name a specific tool or feature for it to work?
A: No. Just name Dobbin and ask—"use Dobbin to pull our objection-handling framework, then draft a reply." Naming Dobbin tells both systems where the context should come from; you don't need to know the mechanics underneath.
• Q: Does Dobbin learn from my Claude conversations?
A: No. In Claude, it's one-way recall. Dobbin learns only from your Slack conversations—nothing you say or do in a Claude chat gets absorbed back into its knowledge. You still get full read access to everything Dobbin knows; the information just travels one way, out to you.
• Q: So does that mean Dobbin knows less in Claude than in Slack?
A: No—recall is complete. You can pull anything in Dobbin's knowledge base, including everything it learned from Slack. The only limit is on writing in, not reading out.
• Q: Will my Claude chats end up visible to my team in Slack?
A: No. Because nothing flows back into Dobbin, your Claude conversation stays yours. It won't surface in Slack, won't train Dobbin, and won't show up for teammates.