MCP Server

HelpYap ships a hosted Model Context Protocol server, so you can connect Claude (or any MCP client) directly to your project. Ask your assistant to review customer conversations, rewrite your bot's prompting, update knowledge base articles, or pull support analytics, all without opening the dashboard.

Endpoint

The server uses the Streamable HTTP transport and authenticates with a HelpYap API key:

https://app.helpyap.com/api/mcp
Authorization: Bearer hy_live_...

Create a key under Dashboard → API access. The scopes you select decide which tools the assistant can see and call, so a read-only key is a safe default: add write scopes only when you want the assistant to make changes.

The server is stateless: there are no sessions and nothing is remembered between calls. Every request carries the key, and every tool call names the project it acts on. That is what lets one endpoint serve every user and every project.

Two kinds of key

KeyReachesUse it when
Personal keyEvery project you are a member of, now and in the future. Write access follows your role in each project (members keep conversation, customer, issue, knowledge, and survey writes; owners and admins keep everything the key grants).You connect Claude for yourself and work across several projects. Each teammate creates their own, so actions are audited under their name.
Project keyOne project, with exactly the scopes it was created with. Owners and admins create them.A shared automation or integration that should only ever see one project.

Picking a project per call

With a personal key, every project-scoped tool accepts a project_id argument (the project id or slug). Call list_projects to see your options; if you belong to a single project the argument is optional. Project keys ignore it (or reject a different project). Assistants handle this themselves: ask "show me open conversations in Acme" and Claude will look up the project and pass it along.

The REST API accepts the same personal keys: send X-HelpYap-Project: <id or slug> (or ?project=) to pick the project.

Connect a client

Claude Code

bash
claude mcp add --transport http helpyap https://app.helpyap.com/api/mcp \
  --header "Authorization: Bearer hy_live_..."

Claude Desktop and claude.ai

Add a custom connector in Settings → Connectors with the URL https://app.helpyap.com/api/mcp and your API key as a Bearer authorization header.

Other MCP clients (Cursor, Windsurf, etc.)

json
{
  "mcpServers": {
    "helpyap": {
      "url": "https://app.helpyap.com/api/mcp",
      "headers": {
        "Authorization": "Bearer hy_live_..."
      }
    }
  }
}

Tools

Every tool except list_projects acts on one project: the key's own project for project keys, or the project_id you pass for personal keys. A tool only appears to the assistant when the key holds its required scope.

ToolRequired scopeDescription
list_projects(none)List the projects this key can act on, with the id and slug to pass as project_id
list_conversationsconversations:readList recent conversations, optionally filtered by status
get_conversationconversations:readRead a conversation with its full message history
reply_to_conversationconversations:writeSend a human reply to the customer (widget, email, SMS, WhatsApp, Messenger)
update_conversationconversations:writeChange status, set tags, or leave an internal note
get_bot_configconfig:readRead the AI agent configuration: prompting, messages, model settings
update_bot_configconfig:writeUpdate prompting: custom AI instructions, welcome message, blocked topics, temperature, and more
get_system_promptconfig:readPreview the fully compiled system prompt the AI agent runs with
list_kb_documentsknowledge:readList knowledge base documents
get_kb_documentknowledge:readRead the full content of a document
create_kb_documentknowledge:writeAdd a new document (embeddings generated automatically)
update_kb_documentknowledge:writeEdit a document's title or content
delete_kb_documentknowledge:writePermanently delete a document
get_support_summaryreports:readSupport analytics: volume, resolution and escalation rates, breakdowns
list_help_center_collectionsknowledge:readList help center collections
create_help_center_collectionknowledge:writeCreate a help center collection
update_help_center_collectionknowledge:writeUpdate a collection's title, slug, order, or published state
list_help_center_articlesknowledge:readList help center articles across all statuses
get_help_center_articleknowledge:readRead an article's full content
create_help_center_articleknowledge:writeWrite a new help center article (draft or published)
update_help_center_articleknowledge:writeEdit, publish, or archive an article
list_announcementsoutreach:readList banners, release notes, and news announcements
create_announcementoutreach:writeCreate an announcement, optionally scheduled with a CTA
update_announcementoutreach:writeEdit, publish, schedule, or archive an announcement
list_guidesoutreach:readList engagement guides (checklists, tours, tooltips, modals)
create_guideoutreach:writeCreate an engagement guide with steps
update_guideoutreach:writeEdit a guide or change its status
list_canned_responsesconversations:readList reply templates
create_canned_responseconversations:writeCreate a reply template with a shortcut
update_canned_responseconversations:writeEdit a reply template
delete_canned_responseconversations:writeDelete a reply template
list_survey_responsessurveys:readRead feedback survey responses with search and filters
list_roadmap_itemsroadmap:readList roadmap / feature request items with votes
create_roadmap_itemroadmap:writeAdd a roadmap item
update_roadmap_itemroadmap:writeUpdate roadmap status, priority, or published state

What to ask your assistant

  • "Which projects can you see? For each one, summarize this week's escalations."
  • "Read my last 20 support conversations and tell me the three most common complaints."
  • "Customers keep asking about refunds and the bot answers badly. Rewrite the custom prompt so it handles refunds correctly, then add a refund policy article to the knowledge base."
  • "Which conversations escalated this week, and why? Reply to any that are still waiting."
  • "Pull the last 30 days of support metrics and summarize how the AI agent is performing."

Security notes

  • Keys are hashed at rest and shown only once at creation. Revoke a key at any time from the dashboard.
  • Every write action (replies, config changes, knowledge base edits) is recorded in the project audit log. Personal keys are attributed to the user who owns them, project keys to the key itself.
  • Personal keys re-check your project memberships and role on every call. Leave a project and the key stops reaching it immediately; nothing is cached server-side.
  • Theme, channel credentials, and billing settings are never readable or writable through MCP.