Skip to content

MCP server

The MCP (Model Context Protocol) server exposes MiaRec data to external AI clients — desktop AI assistants, agent frameworks, or custom integrations that speak MCP. A connected client can query conversation reports, transcripts, and QA metadata through a fixed set of read-only tools; there are no write tools, so an AI client can never modify MiaRec data.

Client access

  • Endpoint: POST /api/mcp on the web portal (JSON-RPC).
  • Authentication: HTTP Basic with a MiaRec user account. The account's role must have the REST API permission — the same permission that gates the REST API (see Sessions and API usage for monitoring).
  • Scoping: every tool call runs as the authenticated user. The client sees exactly the conversations, users, and groups that user's role and access scope allow — nothing more. To scope an AI client to one tenant or one group, create a dedicated user with a suitably restricted role.

API rate limits and per-tenant API request quotas apply to MCP traffic like any other API traffic.

Available tools

The server exposes 12 read-only tools:

Tool Purpose
build_report Build a report — limited to the Calls Summary and Call Details report types.
list_report_types Discover the report types available to build_report.
list_report_columns Discover the columns available for a report type.
list_filters / list_filter_operators / list_filter_choices Discover the filters, operators, and value choices usable in report queries.
get_conversation_transcript Fetch one conversation's transcript.
list_topics List the configured topics.
list_users / list_groups List users and groups (within the caller's scope).
list_evaluation_forms / get_evaluation_form List and fetch scorecards (evaluation forms).

Restricting exposed columns and filters

Administration › System › MCP Server controls how much of the reporting surface MCP clients can see. For each supported report type (Calls Summary and Call Details) the page shows the report's columns in category groups — including the AI-produced custom-field columns — with checkboxes, plus a list of visible filters.

Only the columns and filters selected here are discoverable and usable through the MCP tools. Use this to keep sensitive or noisy columns out of AI clients platform-wide; the setting is global (System tenant), not per tenant.

Operational notes

  • Tool calls consume ordinary web/API capacity and appear in API usage counters; there is no separate MCP usage meter.
  • Since access rides on a normal user account, all the usual controls apply: disable the account to cut off a client, and review authentication activity in the audit trail.