What you can do with the ChatWith MCP server

How these workflows run

These workflows run inside your MCP client: you approve tool calls, ChatWith executes them with your API key and workspace role, and results return as JSON-RPC content. Plans and roles gate individual tools — the same rules the MCP tools spec describes for server-side tools. Every sequence below uses real tools from the ChatWith MCP server.

1. Answer a question from your website knowledge inside your AI client

Example prompt: “What does our refund policy say about digital products?”

• Tools used: ask_assistant with persist=false, so it avoids the monthly quota unless you choose to save the thread
• Plan: Growth or higher
• Result: Reply text plus the sources listing the passages used — no REST round trip.

2. Find the exact page that answers something, without a chat reply

Example prompt: “Return chunks about SSO setup.”

• Tools used: search_knowledge or retrieve_chunks with an assistant_id and a query (max. 200 characters)
• Plan: Growth or higher
• Result: chunk_id, title, url, a snippet and a similarity score.

3. Fix what the assistant could not answer

Example prompt: “Show top unanswered questions for the last 7 days, then add a FAQ paragraph and sync.”

• Tools used: get_top_unanswered_questions → add_text_source or add_website_source → sync_source → get_sync_status
• Plan: Pro for unanswered insights; Growth for sources
• Result: A knowledge gap filled and the sync job status confirmed.

4. Triage conversations that need review

Example prompt: “List conversations in needs_review, open one, mark it answered.”

• Tools used: list_conversations → get_conversation → set_conversation_status (answered)
• Plan: Growth or higher
• Result: Status updated; if a webhook is connected, conversation.closed fires when it is marked answered.

5. Follow up new leads

Example prompt: “List new leads and mark the first one contacted.”

• Tools used: list_leads → update_lead_status (contacted or qualified)
• Plan: Pro or Agency
• Result: Lead status updated in ChatWith. See chatbot lead capture.

6. Launch a new assistant from a prompt

Example prompt: “Create an assistant, attach a website source, allow my domain, publish, and give me the embed snippet.”

• Tools used: create_assistant → add_website_source or add_file_source → set_assistant_domains → publish_assistant → get_embed_snippet
• Plan: Growth or higher
• Result: A published assistant plus the embed script for your site.

7. Check usage before you hit a limit

Example prompt: “How many conversations have we used this month and which pages drive them?”

• Tools used: get_entitlements → get_usage_summary → get_conversations_by_page
• Plan: Growth or higher (insights metrics on Pro+)
• Result: Entitlements, usage counters and a page breakdown before quota errors.

8. Run client accounts (Agency)

Example prompt: “Create a client workspace and connect a signed webhook to our CRM endpoint.”

• Tools used: create_client_workspace → list_workspaces → connect_webhook (HTTPS only; signing secret shown once)
• Plan: Agency
• Result: An isolated client workspace. Each API key is bound to one workspace, so use separate keys per client. Webhook signatures use HMAC-SHA256 per RFC 2104.

What the MCP server does not do

ChatWith MCP does not provide human operator replies, CRM sync beyond webhooks, Slack or Zapier integrations, or billing tools (checkout and subscription management are intentionally excluded). There is no public REST API — only this MCP endpoint over JSON-RPC.

Read the MCP introduction, connect via the MCP server page, or authenticate with API keys and bearer tokens.