Skip to navigation

Data privacy

Some MCP tools, such as get_conversation, retrieve conversation transcripts that may contain personally identifiable information (PII). Before connecting Ada to an AI assistant, disable model training on your chat data so that end-user information is not used to train AI models.

Disable model training

  • ChatGPT: Go to Settings → Data Controls → disable “Improve the model for everyone”.
  • Claude: Go to Settings → Privacy → disable “Allow training”.

If you use a company or enterprise subscription for ChatGPT or Claude, model training is typically disabled by default. Verify this with your IT administrator.

Disable individual tools

Owners can limit tools for the instance, individual team members, or API keys in Config > PLATFORM > MCP server. See Tool permissions for setup.

The server enforces saved permissions across MCP clients. Client-side tool settings do not replace these controls.

Limits of tool permissions

Disabling get_conversation and get_conversations blocks those tools. Other permitted tools can still return end-user text:

  • run_code_tool can return text from inputs or connected systems through results, logs, or errors.
  • Coaching, test cases, and change sets can include end-user text that someone saved in them.

The improvement and glossary conflict guides return built-in instructions. Neither retrieves stored conversations. The glossary conflict guide also includes the scope you supply.

Read-only tools can still return sensitive text. Tool permissions do not redact responses, detect prompt injection, or filter malicious conversations. If your policy prohibits all free text, review every tool you allow.

Removing tool access blocks new calls. It does not remove content already returned to an assistant or stop a call already authorized.

What data flows through MCP

Depending on which tools you use, the following may be transmitted to your chosen MCP client (for example, Claude or ChatGPT):

  • Conversation transcripts
  • Reasoning history
  • End-user inquiry summaries
  • Automated resolution status and reasons
  • Handoff reason codes and reasoning text. The get_conversation tool applies your redaction rules to reasoning text.
  • CSAT scores and feedback
  • Knowledge article content
  • Coaching content
  • Agent configuration (playbooks, tools, custom instructions)