Public Connectors & MCP Integration Guide

5 mins

Dobbin is Company AI, a secure system that captures an organization’s judgment, operating protocols, and knowledge, and surfaces it directly within the tools your team already uses.

Through the Model Context Protocol (MCP), Dobbin connects directly to custom AI clients, allowing Claude, ChatGPT, Microsoft 365 Copilot, and Gemini to access your company’s verified knowledge base safely and securely.

Supported AI Clients

Anthropic Claude (Claude.ai, Claude Desktop, Claude Mobile)

Connector Type: Custom MCP Server
Server URL: https://api.dobbin.ai/mcp
Authentication: OAuth 2.0 (Dynamic Client Registration / PKCE)
Supported Plans: Free, Pro, Max, Team, Enterprise

Setup Instructions:
Open Claude.ai → Settings → Connectors (or Developer).
Click Add Custom Connector.
Enter Name (Dobbin AI) and Server URL (https://api.dobbin.ai/mcp).
Complete the OAuth sign-in prompt to authorize your Dobbin account. Note: Configure custom connectors on Claude Web or Claude Desktop first; Claude Mobile automatically inherits configured remote connectors.

2. OpenAI ChatGPT (ChatGPT Web & Desktop)

Connector Type: Custom App (Developer Mode)
Server URL: https://api.dobbin.ai/mcp
Authentication: OAuth 2.0
Directory Submission Helper: Upload dobbin-bot/scripts/deployment-files/openai/chatgpt-app-submission.json on the OpenAI App Submission form to auto-fill app title, subtitle, description, category, and developer identity metadata.

Setup Instructions:
Open ChatGPT → Settings → Developer Mode (or Apps).
Click Create New App / Add MCP Server.
Enter Server URL (https://api.dobbin.ai/mcp).
Authorize via Dobbin OAuth sign-in.

3. Microsoft 365 Copilot (Copilot Studio)

Connector Type: Copilot Studio Agent Tool
Server URL: https://api.dobbin.ai/mcp
Authentication: OAuth 2.0 → Dynamic Discovery

Setup Instructions:
Open Copilot Studio and select your Agent.
Go to Tools → Add a tool → Model Context Protocol.
Name the tool Dobbin, enter a short description, and paste https://api.dobbin.ai/mcp.
Select OAuth 2.0, then Dynamic discovery.
Create the connection and complete the Dobbin sign-in prompt. Note: Each Copilot user must authorize their own Dobbin connection.

Available MCP Tools

Dobbin exposes ten secure tools via the Model Context Protocol:

Tool Name	Title	Description
`get_knowledge_directory`	Get Knowledge Directory	Browse available knowledge channels, document counts, types, and sampled topics.
`search_internal_knowledge`	Search Internal Knowledge	Hybrid search across company documents and Living Brand Foundation sections with citation links and evidence locators.
`read_internal_evidence`	Read Internal Evidence	Bounded evidence retrieval for specific document sections using signed locator tokens.
`synthesize_internal_knowledge`	Synthesize Internal Knowledge	Synthesizes a comprehensive briefing on internal knowledge relevant to a question.
`evaluate_strategic_alignment`	Evaluate Strategic Alignment	Evaluates proposals or drafts against operating protocols, positioning, and decision frameworks.
`search_conversations`	Search Conversations	Ranked search over Slack thread discussions with dates and permalinks.
`critique_copy`	Critique Copy	Reviews draft customer-facing copy against brand voice, tone, and language filter rules.
`challenge_response`	Challenge Response	Adversarial stress-testing of proposals for risks, blind spots, and failure modes.
`list_channel_documents`	List Channel Documents	Detailed document listing within a specific channel with filtering by type, topic, date, or query.
`feedback`	Report Feedback	Telemetry tool to report whether a tool result was helpful.

Security & Data Privacy Disclosures

Authentication & Authorization

All MCP connections require user authentication via OAuth 2.0 with PKCE.
Access tokens are short-lived and strictly scoped to the user’s authorized Dobbin workspace instance.
Dobbin enforces strict role-based access control (RBAC): users can only search or retrieve documents from channels they have explicit permission to view in Dobbin/Slack.

Data Usage & Model Training

Zero Third-Party Training: Dobbin never uses customer data or telemetry to train public AI models.
Data Encryption: All data in transit is encrypted using TLS 1.3. Data at rest is encrypted using AES-256 / SQLCipher encryption.

Privacy Policy Addendum for MCP Integrations

When connecting Dobbin to third-party AI clients (Claude, ChatGPT, Copilot, Gemini):

Information Processed: Search queries, conversation context, and retrieved document snippets necessary to fulfill tool execution requests.
Third-Party AI Clients: Queries sent through third-party AI clients are processed under the respective user’s or organization’s agreement with that AI client vendor.
Data Logging & Retention: MCP tool handlers persist turn pairs (user_content and truncated assistant_content up to 2,000 characters) via record_mcp_turn_pair for workspace-level usage monitoring, session history, and analytics. Turn records are retained for 90 days by default (or per custom workspace policy).
Data Rights & Account Deletion: Users may revoke MCP OAuth connections at any time via the Dobbin UI or their AI client settings, invalidating active tokens. Workspace administrators or account holders may request complete erasure of turn logs and account data by contacting support@dobbin.ai.