The ChatWith MCP server

What the ChatWith MCP server does

ChatWith runs a remote MCP server: your website assistant, knowledge sources, conversations and leads exposed as MCP tools and resources. Connect from Claude Code, Cursor, Claude custom connectors or any HTTP client that speaks JSON-RPC 2.0. Read the MCP introduction for the protocol model.

Connection details

• Endpoint: https://YOUR_PROJECT.supabase.co/functions/v1/mcp — the exact value for your workspace is shown in Settings → MCP.
• Transport: Streamable HTTP — POST only, with a JSON-RPC 2.0 body. GET returns 405 with "This MCP server uses Streamable HTTP POST", which matches the MCP transport spec.
• Protocol version: 2025-03-26.
• Server: chatwith 0.1.6.

Implemented JSON-RPC methods: initialize, notifications/initialized, ping, tools/list, resources/list, tools/call, resources/read. Prompts are not supported. See JSON-RPC 2.0.

Tools

59 tools are listed in tools/list, grouped by area with minimum plans. Tool definitions follow the MCP tools spec.

Area	Tools	Minimum plan
Assistants	list_assistants, get_assistant, create_assistant, update_assistant, set_assistant_domains (max. 10), publish_assistant, unpublish_assistant, delete_assistant, get_embed_snippet	Growth
Knowledge	list_sources, get_source_documents, add_website_source, add_help_center_source, add_text_source, add_file_source (PDF or document in base64), sync_source, delete_source	Growth
Chat	ask_published_assistant (creates a conversation and counts toward the monthly quota), ask_assistant (also works on unpublished assistants; counts only when persist=true)	Growth
Conversations	list_conversations, get_conversation, set_conversation_status (active, answered, or needs_review), delete_conversation	Growth
Leads	list_leads, update_lead_status (new, contacted, or qualified), delete_lead	Pro
Insights	get_dashboard_metrics, get_conversations_timeseries, get_top_unanswered_questions, get_conversations_by_page, list_analytics_events (metrics accept a rolling window of 1 to 30 days)	Pro
Usage	get_usage_summary, get_entitlements	Growth
Workspace	get_workspace, update_workspace, delete_workspace, list_members, get_profile, update_profile, list_notifications, mark_notifications_read, search_workspace	Growth
Webhooks	list_integrations, connect_webhook, test_webhook, disconnect_webhook	Agency
API keys	list_api_keys (any role), create_api_key, revoke_api_key (owner or admin)	Growth
Retrieval	search_knowledge, retrieve_chunks	Growth
Jobs and audit	list_jobs, get_sync_status, list_audit_logs	Growth
Team	invite_member, change_member_role, remove_member	Agency
Client workspaces	list_workspaces (any role; lists the client workspaces you can access)	Growth
Client workspaces	create_client_workspace	Agency

Resources

9 concrete chatwith:// URIs are returned by resources/list; the other 5 are templated (.../{id}) and are fetched with resources/read once you have an id. Together they cover assistants, sources, conversations, leads, analytics, entitlements and workspace data. See the MCP resources spec.

• chatwith://assistants
• chatwith://assistants/{id}
• chatwith://assistants/{id}/sources
• chatwith://sources/{id}
• chatwith://sources/{id}/documents
• chatwith://conversations
• chatwith://conversations/{id}
• chatwith://leads
• chatwith://analytics
• chatwith://entitlements
• chatwith://workspace
• chatwith://members
• chatwith://notifications
• chatwith://integrations

Connect a client

• Claude Code, Cursor, Claude custom connectors — add a remote MCP server with HTTP transport, the endpoint URL above, and the header Authorization: Bearer cw_live_….
• Any HTTP client — call tools/call with curl, Python or an automation platform exactly as on the ChatWith API page.

Authenticate first with API keys and bearer tokens, then explore MCP use cases and training a chatbot on your website.

Safety in practice

Tool access respects workspace roles and plan gates. Rate limits apply per key (below). Keys are revocable from Settings → API or with revoke_api_key, and list_audit_logs shows MCP activity.

Limit	Value
MCP requests per API key	60/min
Failed auth attempts per IP	30/min
ask_assistant / ask_published_assistant	20/min
Knowledge source creates	15/hour
test_webhook	10/hour
API key creation	10/hour