Help Center
Clear, practical guides for writing, compiling, and sharing LaTeX and Typst projects.
Connect an external Agent
Let a compatible MCP Agent work in only the projects and areas you choose.
Connect an app by signing in
Apps that add an MCP server by its URL alone, such as Claude, need no token: they send you to JointTex to sign in and approve.
How to do it
- In a project, open External connections in the Activity Bar and choose AI Agent (MCP): it shows the MCP server URL, the project ID and the setup for your app. The same URL is also under Access tokens in Account settings.
- Paste it where the app adds a server. In Claude: Customize, Connectors, Add custom connector.
- The app opens JointTex in your browser. Sign in if asked, check the address the page says you will return to, choose the scopes and projects, and choose Allow. The page starts with the scopes the app asked for, or the three read scopes if it asked for none, and with All projects ticked.
- The app appears under Connected apps in Access tokens, where you can edit its scopes and projects or disconnect it.
Create an MCP token
JointTex Education, Plus, and Pro can connect compatible external Agents, including Codex and Claude Code, through the MCP server shown in Account settings.
How to do it
- In Account settings, open Access tokens, expand Use an MCP token and create one, selecting only the projects and permissions the Agent needs.
- Copy the token once and configure your Agent with the displayed MCP server URL and that token as an Authorization Bearer header.
- Or choose Copy Agent setup prompt and paste it into your Agent. The prompt contains the endpoint and the complete token, so send it only to an Agent you trust.
- Start with read-only scopes. If the Agent later has to change source, use Edit access to add file-write to the existing token.
Choose scopes and projects
A token carries a list of scopes and a list of projects, and the Agent can do nothing outside either. Only the help guides, which are public, can be read with any token.
How to do it
- The scopes are projects:read, projects:create, projects:organize, projects:archive, projects:trash, collections:read, collections:write, members:invite, files:read, files:write, review:decide, compile:run, comments:read, comments:write, templates:publish, account:read, notes:read, notes:write, and references:read. A new token starts with the three read scopes, and a write scope always brings its read scope with it. compile:run brings files:read, because what a compile hands back (its log, a Typst query or evaluation) can show the project's files.
- With projects:create an Agent can create a project in your account, and with members:invite it can invite someone by email to a project you own; both count against your plan's project and collaborator limits. Neither is ticked on a new token or given to an existing one: tick them only when you want the Agent to do this for you.
- collections:read/write read and manage your personal collection tree without selecting projects. projects:organize files selected projects into collections; projects:archive changes only your personal archive. projects:trash lets the owner move selected projects to the recycle bin or restore them, affecting all collaborators. These scopes are off by default and never added to existing connections automatically. Deleting a collection also deletes its subcollections and filing, but preserves the projects; organization changes take effect immediately, outside tracked source changes.
- With templates:publish an Agent can publish a project you own to the template library, or republish a template you published, as Publish as template does; the template is public at once. It is not ticked on a new token or given to an existing one.
- With account:read an Agent can see what your account has set up, so it can help you set up the rest: your plan, whether Zotero, WebDAV and GitHub are connected, how many Git and MCP tokens you hold, your Agent connectors and organizations. It never sees a key, password or token. It is not ticked on a new token.
- With references:read an Agent can search your references as the citation browser does: your personal reference library, the .bib files of the projects the token was given (this also needs files:read), and your connected Zotero library, and it can look a paper up online. It only reads; it is not ticked on a new token, and a token needs no project selected for it.
- With comments:write an Agent can start a review thread on a passage, reply to one, and resolve or reopen it; it cannot edit comments, and it can delete only a thread it started through the same connection.
- What an Agent writes lands as tracked changes for the people in the project to review, as the in-app Agent's edits do. With review:decide it can also accept or reject tracked changes; leave it off if deciding should stay with people.
- With files:write an Agent can also change the project's compile settings (root file, engine, TeX Live version, compile mode, and the inputs a Typst document reads), which apply to every collaborator. With files:read it can get a link that downloads the compiled PDF; the link works for ten minutes without signing in, and stops working when you disconnect the Agent.
- With compile:run an Agent can compile a project, query or evaluate a Typst document, and export the project as Word, Markdown, HTML or an arXiv package. Each compile counts against your own compile capacity. A connection saved with compile:run before it brought files:read keeps what it had, and cannot export, until you save it in Edit access.
- With notes:read an Agent can list, search and read your Personal Notes, and with notes:write it can also create, edit, move and delete them. Notes belong to your account, so these scopes need no project selected; neither is ticked on a new token.
- Choosing All projects includes projects you can access now and in the future, so name the projects when the Agent's job is narrow.
- A token never exceeds your own role: in a project where you are a viewer, the Agent cannot write even with files:write.
Watch and revoke the connection
A token you cannot see is a token you cannot govern, so every connection stays listed and reversible.
How to do it
- Review the token under Use an MCP token in Access tokens to see which projects and scopes it may reach.
- Use Edit access to add or remove scopes and change projects when the work changes. Saved changes apply on the next call with the existing token.
- Revoke it the moment it is no longer needed; revocation takes effect immediately.
- Whatever an external Agent writes appears in the project's history like any other edit, so you can review or restore it.
- You can keep up to 10 MCP tokens. If your plan drops below Education, Plus, or Pro, existing tokens stay saved but are paused.
