Point Claude, ChatGPT, Cursor, Claude Code or Replit Agent at Helpdesky and it can write, organize and publish help center articles as you — within your operator permissions. Sign in and approve. Nothing to install, no keys to paste.
On it — reading what you already have.
search_help_centerquery: "refund" → 3 articlesget_article× 3 — full Markdownlist_categories→ Billing (cat_7f2)create_articletitle: "Refunds FAQ", category: Billing, 9 questionspublish_article→ live, embeddings generatedDone. I combined the refund window, partial refunds and the annual-plan exception into nine questions, each linking back to the original article.
118 tools, generated from the same code the dashboard uses: 51 read-only, 50 that write, 17 flagged destructive. Every call goes through the same permission checks and plan limits as a click in the dashboard.
Tools come in tool sets. A new connection gets the Support desk preset (59 tools: inbox, knowledge base, contacts, team); ask the agent for setup or growth work and it enables the Setup (41) or Growth (18) sets itself — or pick them on the consent screen or in the dashboard. Clients with a tool cap use /mcp/x/lite (33 tools).
Draft in Markdown, update titles, excerpts and slugs, publish when ready or pull an article back to draft. Published articles appear in search, the widget and Ask AI.
create_articleupdate_articlepublish_articleunpublish_articleCreate and rename categories, reorder them, move articles between them, and read the whole structure to spot gaps.
list_categoriescreate_categoryupdate_categoryreorder_categoriesSearch published articles, read any of them in full, or ask your help center's Ask AI the way a visitor would and get the answer with its sources.
search_help_centerget_articleask_help_centerRenaming a slug creates a redirect by default; the agent can list, retarget or remove redirects so old links keep landing.
list_redirectsupdate_redirectdelete_redirectThe agent lists the helpdesks you belong to and targets one explicitly. It never guesses: a helpdesk you cannot access returns a clear error.
list_helpdesksget_helpdeskget_profileThe search and fetch tools follow OpenAI's connector contract, so ChatGPT can cite your published articles in its answers.
searchfetchFull list with descriptions: /docs/mcp · machine-readable: /mcp/catalog.json
Is your agent also writing the embed, proxy or template code in your own app? Give it the AI Agent Skill as well: the MCP server operates your dashboard, the skill covers the code.
Plain language in, published help center out. Each prompt below only uses tools that ship today.
No API keys to paste for Claude or ChatGPT. The agent acts as you, and only as you.
OAuth 2.1 for hosted clients. Paste the endpoint URL into Claude or ChatGPT; the client registers itself (dynamic client registration), sends you to Helpdesky to sign in, and shows a consent screen. Approve it and the client receives short-lived access tokens with rotating refresh tokens. PKCE is required.
Personal tokens for tools that take a header. Cursor, Claude Code, scripts and custom agents can use a personal MCP token created in Dashboard → MCP, sent as Authorization: Bearer hdmcp_pt_…. A token is shown once and stored hashed.
Every approval and token is listed on the same page with when it was last used. Disconnect an agent or revoke a token and it stops with the next call.
destructiveHint, and agents are instructed to confirm with you first.The endpoint is the same everywhere: https://helpdesky.io/mcp. Only where you paste it changes.
https://helpdesky.io/mcp and save.https://helpdesky.io/mcp and choose OAuth as the authentication.search and fetch let ChatGPT cite your articles..cursor/mcp.json:{ "mcpServers": { "helpdesky": {
"url": "https://helpdesky.io/mcp/x/lite",
"headers": { "Authorization": "Bearer hdmcp_pt_…" } } } }
/x/lite keeps Helpdesky under Cursor's 40-tool cap; use https://helpdesky.io/mcp/x/all for everything.claude mcp add --transport http helpdesky \ https://helpdesky.io/mcp
/mcp inside Claude Code and sign in when it opens the browser — or append --header "Authorization: Bearer hdmcp_pt_…" to the command above to use a token instead.https://helpdesky.io/mcp as a remote MCP server (Streamable HTTP) wherever your agent configures tools./.well-known/oauth-protected-resource/mcp.initialize and tools/list work, so a client can inspect the tools before you sign in.Exact menus move between client versions; /docs/mcp is kept current, and the public page at https://helpdesky.io/mcp shows the same instructions.
Everything a listing form or a connector review asks for, in one place.
| Server name | Helpdesky |
|---|---|
| Registry name | io.helpdesky/helpdesky — the name reserved for the official MCP Registry (server.json with a streamable-http remote, domain verified at /.well-known/mcp-registry-auth). Until the entry is published, connect with the endpoint URL. |
| ChatGPT | Packaged as a ChatGPT plugin (Agent Plugins manifest referencing this MCP connection plus a Helpdesky skill); the directory listing is pending review. Until it is live, connect in developer mode with the endpoint above — search/fetch already let ChatGPT cite your articles. |
| Endpoint URL | https://helpdesky.io/mcp |
| Transport | Streamable HTTP (stateless, JSON responses). No stdio package to install. |
| Authentication | OAuth 2.1 — authorization code with PKCE (S256), dynamic client registration (RFC 7591), authorization server metadata (RFC 8414), protected resource metadata (RFC 9728), token revocation — or a personal token as Authorization: Bearer hdmcp_pt_…. Authorization server metadata · Protected resource metadata |
| Tool catalog | https://helpdesky.io/mcp/catalog.json — 118 tools with JSON Schema input/output and annotations (list_tool_sets, enable_tool_sets, get_profile, list_helpdesks, get_helpdesk, list_articles, get_article, create_article, update_article, publish_article, unpublish_article, delete_article, list_categories, create_category, update_category, reorder_categories, delete_category, list_redirects, update_redirect, delete_redirect, search_help_center, ask_help_center, search, fetch, list_conversations, get_conversation_counts, get_conversation, list_conversation_messages, reply_to_conversation, edit_message, delete_message, translate_message, set_conversation_status, assign_conversation, mark_conversation_read, set_conversation_contact, set_conversation_notification_emails, mass_close_conversations, delete_conversation, list_contacts, list_contact_conversations, block_contact, unblock_contact, update_contact_email, list_operators, list_assignable_operators, invite_operator, invite_operator_batch, update_operator_permissions, update_operator_role, update_operator_notifications, remove_operator, get_notification_settings, update_notification_settings, get_messaging_settings, update_messaging_settings, get_automation_settings, update_automation_settings, get_helpdesk_settings, update_helpdesk_settings, update_branding, update_seo_metadata, detect_branding, check_slug_availability, change_helpdesk_address, create_helpdesk, get_hosting_status, get_custom_domain_status, set_custom_domain, remove_custom_domain, save_subfolder_hosting, check_subfolder_setup, activate_subfolder_hosting, remove_subfolder_hosting, forget_previous_subfolder_address, save_headless_hosting, check_headless_setup, activate_headless_hosting, remove_headless_hosting, forget_previous_headless_address, get_hosting_instructions, get_widget_settings, update_widget_settings, get_ticket_center_settings, update_ticket_center_settings, get_contact_form_settings, update_contact_form_settings, get_embed_snippets, get_ai_settings, update_ai_settings, check_ai_configuration, list_ai_models, list_ai_questions, list_data_sources, add_data_sources, update_data_source, delete_data_source, rescrape_data_source, discover_data_source_urls, get_seo_suggestions, preview_seo_fix, apply_seo_fix, preview_internal_links, add_internal_links, preview_seo_run, run_seo_optimization, list_seo_changes, undo_seo_change, get_seo_automation, update_seo_automation, start_link_scan, get_link_scan_results, fix_link, update_link_follow, get_link_scanner_settings, update_link_scanner_settings, get_job_status, get_overview) |
| Documentation | https://helpdesky.io/docs/mcp |
| Who can use it | Anyone with a Helpdesky account, on every plan including Free. The agent acts within the connected operator's permissions and plan limits. |
| Data accessed | Only for the helpdesks the connected user belongs to and only within their permissions and enabled tool sets: help center content (articles incl. drafts, categories, redirects), the user's profile (name, email, role), team members (names, emails, roles, permissions), customer conversations and contacts (names, email addresses, full message text incl. internal notes), the help-center AI question log, and helpdesk settings. Never billing, passwords, API keys, MCP tokens or the ticket-center signing secret; provider API keys are write-only. Details in the privacy policy. |
| Support contact | support@helpdesky.io · https://helpdesky.io/contact |
| Privacy policy | https://helpdesky.io/privacy |
| Terms of service | https://helpdesky.io/terms |
| Publisher | Helpdesky, built by Massive Monkey Software — helpdesky.io |
Free to start. Connect in a minute, disconnect in a click.