MCP Integration

SupportBot ships with a Model Context Protocol (MCP) server: Claude and any other MCP-capable AI client can operate your support platform directly through 44 tools — inspect conversations, upload knowledge-base documents, run test scenarios, trigger weekly reports, and more.

Endpoint

Field	Value
URL	`https://showcase.<your-domain>/api/mcp`
Transport	Streamable HTTP (MCP standard; `POST`/`GET`/`DELETE`)
Authentication	`Authorization: Bearer <API key>`
Rate limit	60 requests per minute per key
Sessions	Stateful; idle sessions close after 30 minutes

Creating an API key

Open Settings → API Keys in the admin panel.
Give the key a name (e.g. "Claude Desktop") and create it.
The key, starting with sk_showcase_, is shown only once — store it somewhere safe. The server keeps a SHA-256 digest, never the key itself.
Revoke unused keys from the same screen. Every MCP call is attributed to the calling key and user (audit trail).

Connecting Claude

Claude Code (CLI):

claude mcp add supportbot \
  --transport http https://showcase.<your-domain>/api/mcp \
  --header "Authorization: Bearer sk_showcase_..."

Claude Desktop / other clients: add the same URL as a Streamable HTTP MCP server and set your key in the Authorization header.

To try the connection from a browser, use the /mcp-test page in the showcase app.

Tool catalog

Tools carry MCP behavior annotations (read-only / mutating / reaching external systems), so your client can present confirmation flows accordingly.

Monitoring & conversations (read-only)

Tool	Purpose
`get_overview`	Overview metrics (conversation counts, resolution rate)
`get_recent_conversations`	Recent conversations; topic/sentiment/time filters
`get_conversation`	Single conversation detail
`get_quality_report`	Bot quality report
`get_system_health`	Service health (Chatwoot, Dify, DB, Redis)

Knowledge base

Tool	Purpose
`list_datasets` / `list_documents`	List datasets and documents
`upload_document`	Upload a document to the knowledge base
`get_document_status`	Track indexing status
`delete_document`	Delete a document
`query_knowledge`	Search the knowledge base (RAG query)
`get_knowledge_stats`	KB statistics

Configuration & appearance

Tool	Purpose
`get_config` / `update_config`	Bot configuration (name, welcome message, LLM model…)
`get_appearance` / `set_appearance`	Widget colors and appearance
`get_widget_config` / `update_widget_config`	Widget configuration
`get_widget_snippet`	Embed code for your site
`list_topics` / `create_topic` / `update_topic` / `delete_topic`	Topic taxonomy (delete = archive)

Personas & testing

Tool	Purpose
`list_personas` / `list_customer_personas`	Bot and customer personas
`create_bot_persona` / `update_bot_persona` / `delete_bot_persona`	Persona management
`set_default_persona`	Change the default persona
`list_scenarios` / `run_scenario`	List/run test scenarios
`run_acceptance_suite`	Run the acceptance test suite
`run_rag_eval`	End-to-end RAG validation test
`run_rag_metrics`	Ragas-style RAG metrics (faithfulness, context precision/recall…)
`send_test_message` / `test_model` / `classify_message`	One-off test calls

Reports & operations

Tool	Purpose
`list_reports` / `get_report`	List/read weekly reports
`trigger_weekly_report`	Trigger report generation manually
`submit_conversation_feedback`	Attach operator feedback to a conversation
`list_webhook_events` / `retry_webhook_event`	Inspect/retry the webhook queue
`list_api_keys`	List API keys

Security notes

The endpoint validates the Origin header on browser-originated requests (DNS-rebinding protection). Non-browser clients such as Claude Desktop/CLI are unaffected; extra trusted browser origins go in the MCP_ALLOWED_ORIGINS environment variable.
Keys are stored as irreversible digests; if you suspect a leak, revoke the key in the panel and mint a new one.
Mutating tools (configuration, documents, topics) are written to the audit log.