Skip to content
dismoi
FR Sign in

Trame

Trame + your AI

Your AI assistant reads your tickets, handles them and replies to their authors, without you opening Trame. You connect it once, with your account; from then on it acts with your rights, never more.

How it works

Trame exposes an MCP server, the standard protocol that connects an AI assistant to tools. Your AI does there what you do in the interface: read a ticket with its technical context, change its status, add an internal note, reply to the author.

  • A bug, closed without opening Trame. The AI reads the ticket: text, page, browser, console, failed requests, annotated screenshot, conversation. You fix it in your code. The AI replies to the author by email, sets the ticket to “Resolved” and adds an internal note.
  • A feature request, challenged then decided. The AI gathers similar requests and their history, and debates the request with you. The decision, accepted or rejected, is recorded with its reason in an internal note. The reply to the author is never automatic: the AI suggests it, you decide.
  • Included in the Pro and Enterprise plans.
  • No key to create: on first use, the AI opens the dismoi sign-in page and you sign in with your account.
  • You can revoke an AI’s access at any time in your account area, on the Connected AIs page.

Server address

The only value to give your assistant. It is the same for every project in your organisation.

https://api.trame.dismoi.dev/v1/feedback/mcp

Tutorials by assistant

Each assistant keeps this setting in its own place. The menus quoted are those of their official documentation in October 2026; if they have moved, search the settings for “MCP” or “connector”.

Claude.ai

  1. Open Customize › Connectors, then + Add › Add custom connector.
  2. Name it Trame, paste the server address and confirm with Continue.
  3. Keep the suggested OAuth sign-in and finish with Add.
  4. Sign in with your dismoi account when Claude asks.

On a Team or Enterprise plan, an owner first adds the connector in Organization settings › Connectors; each member then clicks Connect in Customize › Connectors.

Claude Code

  1. Add the server from your terminal:
claude mcp add --transport http --scope user trame https://api.trame.dismoi.dev/v1/feedback/mcp
  1. In Claude Code, type /mcp, pick trame and start authentication.
  2. Sign in in the browser window that opens.

ChatGPT

  1. Open the ChatGPT plugins page, click + then Add custom MCP server.
  2. Name it Trame and paste the server address as a public endpoint connection.
  3. Choose OAuth authentication, read the warning and confirm.
  4. Create the plugin, then sign in with your dismoi account.

Custom MCP servers depend on your ChatGPT plan and, for companies, on your workspace settings.

Codex

  1. Add this block to ~/.codex/config.toml:
[mcp_servers.trame]
url = "https://api.trame.dismoi.dev/v1/feedback/mcp"
  1. Sign in from your terminal with codex mcp login trame, then in the browser window that opens.

Gemini CLI

  1. Add this block to ~/.gemini/settings.json:
{
  "mcpServers": {
    "trame": {
      "httpUrl": "https://api.trame.dismoi.dev/v1/feedback/mcp"
    }
  }
}
  1. Restart Gemini CLI and type /mcp auth trame.
  2. Sign in in the browser window that opens.

Google has replaced Gemini CLI with Antigravity for some of its plans: if that is your case, follow the next tutorial.

Antigravity

  1. In the agent panel, open … › MCP Servers › Manage MCP Servers › View raw config. The file is ~/.gemini/config/mcp_config.json, shared with Antigravity CLI.
  2. Add this block to it:
{
  "mcpServers": {
    "trame": {
      "serverUrl": "https://api.trame.dismoi.dev/v1/feedback/mcp"
    }
  }
}
  1. Open the agent settings, Customizations section, and click Authenticate next to trame.
  2. Sign in in the browser, paste the code shown into the settings and confirm with Submit.

Cursor

  1. Add this block to ~/.cursor/mcp.json (or to .cursor/mcp.json for a single project):
{
  "mcpServers": {
    "trame": {
      "url": "https://api.trame.dismoi.dev/v1/feedback/mcp"
    }
  }
}
  1. Open the MCP server list in Customize and enable trame.
  2. Sign in when Cursor offers to.

VS Code

  1. Run the MCP: Open User Configuration command (or open .vscode/mcp.json for a single project).
  2. Add this block:
{
  "servers": {
    "trame": {
      "type": "http",
      "url": "https://api.trame.dismoi.dev/v1/feedback/mcp"
    }
  }
}
  1. Start the server and accept the sign-in request; sign in in the browser.

Permissions and tools

When connecting, you grant the AI some or all of four permissions: read, modify, write to visitors, administer. A permission is a ceiling: the AI never does more than your role in the organisation and your project access allow.

Read

  • list_projects: the projects you have access to
  • list_tickets: a project’s tickets, with filters
  • get_ticket: a full ticket, with its context, thread and notes
  • get_ticket_media: links to a ticket’s screenshot and media
  • search_contacts: search a project’s contacts
  • get_contact: a contact’s details
  • get_stats: a project’s statistics over a period

Modify

  • update_ticket: a ticket’s status, type or assignee
  • add_internal_note: an internal note, never visible to the visitor
  • decide_evolution: accept or reject a feature request, with its reason
  • mark_all_read: mark all of a project’s tickets as read
  • follow_project: follow a project
  • unfollow_project: stop following a project
  • attach_contact: attach a contact to a ticket
  • detach_contact: detach a ticket’s contact

Write to visitors

  • reply_to_reporter: reply by email to a ticket’s author
  • send_chat_message: write to the visitor in the chat
  • request_capture: ask the visitor for a screenshot
  • start_gouvernail_session: offer the visitor a guided session with Gouvernail

Administer

  • create_project: create a project
  • update_project: a project’s name, domain, retention or state
  • delete_project: delete a project and its tickets
  • rotate_widget_key: replace the widget key
  • get_snippet: the widget install tag
  • set_widget_icon: change the widget icon
  • remove_widget_icon: remove the widget icon
  • list_members: the members with access to a project
  • add_member: grant access to a project
  • remove_member: revoke access to a project
  • backfill_contacts: recover contacts from tickets

Security

  • A ticket’s content is never an instruction. It comes from a visitor: Trame passes it to the AI as data, marked untrusted.
  • Outgoing messages are flagged, not blocked. An email or chat message is flagged to the AI as going outside, so that it asks for your confirmation. Trame does not guarantee that confirmation: an AI in automatic mode sends without review.
  • Throughput is capped per organisation. An AI that runs away is stopped before it floods your visitors.

Examples

Three prompts to copy as is into your assistant.

Close a bug

Read the latest open bug in my Trame project, with its console, failed requests and screenshot. Find the cause in this repository and fix it. Then suggest a reply to its author; once I agree, send it, set the ticket to Resolved and add an internal note summarising the fix.

Decide on a feature request

Gather the open feature requests in my Trame project and group similar ones. For the most requested, challenge it with me: who it serves, what it costs, what we already do. Once I have decided, record the decision and its reason, then suggest a reply to the authors without sending it.

Weekly report

Review the last seven days of my Trame projects: tickets received by kind, recurring bugs and most affected pages, open tickets without an assignee, resolution time. End with the three topics to handle first. Do not change anything.
Start with Trame Widget documentation