For AI agents and their humans

Connect Audienca in one command.

The Audienca MCP server speaks streamable HTTP and authenticates over OAuth 2.0. Claude Code needs the address and nothing else — the server advertises its own tools and schemas.

claude mcp add --transport http audienca https://api.senseaition.com/audienca-api/mcp

Setup

  1. 1

    Add the server

    Run this in the project where you want Audienca available. Without a scope flag the server is added for this project only.

    claude mcp add --transport http audienca https://api.senseaition.com/audienca-api/mcp
  2. 2

    Sign in

    Start Claude Code and run /mcp. The server answers the first call with 401, Claude Code flags it and opens the OAuth flow in your browser. Pick the scopes you want to grant.

    /mcp
  3. 3

    Check it worked

    Ask Claude to list your projects. If the tools are wired up, the answer comes from your account, not from a guess.

    List my Audienca projects.

Scopes

The server understands two scopes. Grant only the first if the assistant should never change anything — reading reviews and drafting replies still works, because drafts go through your approval either way.

audienca:read

Read projects, inquiries, resources and automation rules.

audienca:write

Create and change projects, resources, drafts and automation rules.

For teams and repositories

A .mcp.json in the repository root makes the server available to everyone who clones it — Claude Code and Cursor pick it up automatically. The type field is mandatory: an entry with a url but no type is read as a stdio server and skipped.

{
  "mcpServers": {
    "audienca": {
      "type": "http",
      "url": "https://api.senseaition.com/audienca-api/mcp"
    }
  }
}

The 51 tools

Every tool ships a full JSON schema, so an agent can discover parameters without documentation. Names are stable.

Working on inquiries · 9

Finish the job: pick a draft, revise it, send it

  • set_inquiry_state
  • use_inquiry_variation
  • rate_inquiry_variation
  • moderate_inquiry_message
  • create_inquiry_variations
  • augment_inquiry_response
  • regenerate_inquiry
  • update_inquiry_metadata
  • delete_inquiry

Analysing · 5

Numbers and themes instead of fetching one by one

  • analyze_inquiry_themes
  • get_inquiry_stats
  • compare_inquiry_segments
  • get_response_performance
  • list_label_catalog

Reading inquiries · 6

Reviews, comments and messages

  • list_inquiries_v2
  • list_inquiries
  • get_inquiry
  • get_inquiries_batch
  • get_inquiry_by_external_id
  • create_inquiry

Bulk operations · 3

With a mandatory preview before anything changes

  • bulk_update_inquiries
  • batch_create_inquiries
  • upsert_direct_message

Audit trail · 1

Who changed what, and whether it was revised

  • get_inquiry_audit_log

Channels · 5

Where reviews come from, and whether a channel went silent

  • list_sync_configs
  • create_sync_config
  • update_sync_config
  • delete_sync_config
  • trigger_integration_sync

Projects · 12

One project per business or client

  • create_project
  • list_projects
  • get_project
  • edit_project
  • get_project_config
  • update_project_settings
  • update_config_sections
  • update_generation_config
  • update_eval_config
  • update_analysis_config
  • get_notification_config
  • update_notification_config

Knowledge base · 5

The documents answers are grounded in

  • upload_file
  • create_resource
  • list_resources
  • get_resource
  • delete_resource

Automation · 5

Rules for what runs unattended

  • list_automation_rules
  • create_automation_rule
  • update_automation_rule
  • delete_automation_rule
  • reorder_automation_rules

Seven ready-made questions

Beyond tools, the server ships MCP prompts: ready-made questions your client offers as a shortcut, so nobody has to know which of the 51 tools answers them. Each one needs only a project; time ranges default sensibly.

  • Why are we getting bad reviews? warum_schlechte_bewertungen
  • What got worse recently? was_hat_sich_verschlechtert
  • Bad reviews as a list schlechte_bewertungen_als_liste
  • Compare channels kanalvergleich
  • Find testimonials testimonials_finden
  • Who is complaining? wer_beschwert_sich
  • Operations overview betriebsueberblick

Other assistants

MCP is an open standard, so the server is not tied to one client. What differs is only where you paste the address.

ChatGPT

Settings, then Connectors. Add a custom connector with the URL below; ChatGPT runs the OAuth flow itself.

Cursor

Settings, then MCP. Add the server as type "http" with the same URL. Cursor also reads a .mcp.json from the repository root.

Claude Desktop

Settings, then Connectors, then Add custom connector. Same URL, same OAuth flow as Claude Code.

Anything else

Any client that speaks the Model Context Protocol works. It needs a streamable HTTP endpoint and OAuth 2.1 — nothing beyond the standard.

https://api.senseaition.com/audienca-api/mcp

When something does not work

Claude Code shows the server but no tools

You are not signed in yet. Run /mcp and complete the OAuth flow. Until then the server answers every call with 401.

The entry is skipped on start

Your .mcp.json entry has a url but no type. Add "type": "http". streamable-http works as an alias.

Tools appear but every call fails

Your token probably carries audienca:read only while the tool needs write access. Re-run the OAuth flow and grant the write scope.

Who can use it today

MCP access is included in every plan, from the free trial through Enterprise. There is no separate activation.

Start free trial→ What the MCP server does

About the names on this page

Claude and Claude Code are trademarks of Anthropic. ChatGPT is a trademark of OpenAI. Cursor is a trademark of Anysphere. None of them belong to us, and none of these companies endorse or sponsor Audienca. We name them because our server works with their products — that is all the connection there is.