The Gleap MCP server for your AI tools.

The Gleap MCP server is a hosted Model Context Protocol server at https://mcp.gleap.ai/mcp (https://mcp.us.gleap.ai/mcp for projects hosted in the US) that connects an AI client to one Gleap project. It exposes over 90 tools for tickets, conversations, contacts, companies, help center, outbound messages (surveys, emails, chat messages, banners, info cards, checklists, tours, tooltips, push and news), feature requests, statistics, pipelines, workflows and coding sessions, so an assistant can read customer context and take supported actions with your review.

Any MCP client that supports Streamable HTTP and OAuth: Claude.ai, Claude Desktop and Claude Code, Codex in the CLI and IDE extension, Cursor, and custom clients. Clients that can send custom headers may use an API key instead of OAuth. The setup guide lists the current client steps.

OAuth with PKCE and dynamic client registration. Your client opens the Gleap sign-in, you choose the project, and Gleap issues a token scoped to that project. No API key appears in a prompt or a deep link. Clients that support custom headers can send x-gleap-api-key and x-gleap-project-id instead.

Yes. It is an MCP server built around a customer support workspace: tickets and their full conversations, the customer and company behind them, the help center they were answered from, the feature requests they voted on and the statistics of the support team. It also covers pipelines, workflows and the coding sessions that fix confirmed bugs.

Read-only tools include search_tickets, find_tickets, get_ticket_messages, search_contacts, get_help_center_article, search_feature_requests, get_statistics_facts, search_knowledgebase and ask_code. Writes such as send_message, update_ticket, create_help_center_article and run_ticket_workflow are annotated as writes so your client can ask before running them.

No. A coding agent working a ticket shows you each customer reply in its own conversation and sends it with send_message only after you say yes; asking it to answer the customer counts as that yes. Nothing is staged in the ticket composer. Only the two fix messages that come with a pull request skip that step: Gleap sends the first right away and the second when the fix is live, following your settings. Your client's tool approvals still decide which other calls need your confirmation.

On plans with Kai Code, opening a ticket in Claude Code, Codex or Cursor from Gleap creates a coding session and starts the tool with a kickoff prompt. The tool fetches the ticket over MCP and reports every step with report_progress: what it is investigating, a question when it needs an answer, the pull request when it opens one. The ticket shows the status live. With the PR the tool passes two customer messages: Gleap sends one right away (the fix is built and is being reviewed) and the other when the fix is live, following your post-merge setting. The session owns the ticket until it is done: the customer confirms the fix, or 24 hours pass without a reply, and a merge never closes it. A teammate who takes the ticket over or replies to the customer stops the session. Read more about Kai Code.

Yes. Add MCP servers in Kai Support › Tools, from presets such as Gleap, Linear, GitHub, HubSpot, Jira and Stripe to any custom HTTP or stdio server, and set each one to always or only on hard tickets. Kai Code loads the same servers. Custom agents pick tools from the same catalog, with individual tools switched off per agent.

During an investigation Kai Support may query the connected MCP servers, for example a database, to answer questions the customer would otherwise have to re-supply. Money and deletes, such as refunds, plan changes and deleting data, wait for a person's OK on the card before Kai Support acts. Its own Gleap access during an investigation is five tools: the customer's ticket history, similar tickets and feature requests. See Kai Support.

Use the Server API when your own code calls Gleap, for imports and application integrations. Use MCP when an AI client chooses and calls tools in response to a request. The integration directory covers native integrations and Composio connectors.

No. The Gleap MCP server is part of the platform on Team, Pro and Enterprise; Starter has no API or MCP access. Usage in your own AI client is billed by that client. Kai Support's investigations, Kai Code and custom agents are included from Team, with AI usage billed separately. See pricing.

Tokens are scoped to one project and verified on every request. Tool annotations identify write operations for your client. Coding agents ask you before a message reaches a customer, and a stopped session can no longer write; configure client approvals and agent permissions for the other actions you allow. The trust center covers SOC 2, GDPR and the security overview.

Yes, the settings you are allowed to change yourself, such as opening hours, SLAs, teams and the widget. It reads the settings page first, and risky changes need your confirmation. SSO, the sending domain, channel connections, API keys and users stay in the dashboard. Kai Copilot uses the same rules when you ask it inside Gleap.

Yes. Agents draft surveys, emails, chat messages, banners, info cards, checklists, product tours, tooltips, push and news, then set the trigger, page rules, audience and schedule. get_targeting_options lists the events and contact attributes your project tracks, and preview_outbound_audience counts who matches before anything goes live. When the audience needs an event you do not track yet, get_event_tracking_guide returns the tracking code for your SDK or backend. Publishing is a separate tool for after your approval; one-off sends need the recipient count you approved, and one-off emails and push never go to every contact. See outbound messages.

The Gleap platform keeps conversations, bug reports, feature requests and knowledge in one workspace. MCP extends that context to the AI tools your team already uses and lets coding tools close the loop on tickets. Read how Gleap agents close the loop from report to release.