Use Sprintree from Claude, ChatGPT or any MCP client

Pages

Pages are a project's documents: a tree of titled pages, each with a body written in the same editor as an issue description. Anyone who can read the project can read its pages; editors and owners can write them. A project holds up to 2000 pages, nested up to 10 levels deep. Pages are numbered per project and live at /p/KEY/pages/N. The page tools are always listed; their arguments are on each tool's page.

• list_pages: the top level, the children of a page (parent), a whole subtree flattened in tree order (depth: "all", 500 pages at most) or a text search (query). The rows (number, title, parent, child count) are under nodes, paged with cursor (a missing cursor means the last page). Like the other read tools it answers compact JSON; format="text" gives the readable Markdown.
• get_page: title, breadcrumbs, link, base_rev and the body as Markdown (a JSON document by default, format="text" for the readable Markdown page). Optionally the pages and issues that link to it (include: ["backlinks"]), its saved versions, or an old version's body (version). A body over 64 KiB is cut and the answer says so in note: do not send a cut body back with update_page; edit such a page in the web app.
• create_page, update_page, move_page (under another parent, or just reordered) and delete_page. Deleting takes the subpages along and answers a batch id; call delete_page with restore set to that id to undo it.
• get_issue_details with include: ["pages"] lists the pages that mention an issue.

Comments and the inbox

• add_comment takes project and page (the page number) instead of an issue key. reply_to, quote (an inline thread, matched against the page body) and idempotency_key work as on an issue.
• get_page with include: ["comments"] returns the threads, with the ids that add_comment (reply_to), resolve_thread and reopen_thread take; the last two answer a page thread with a page field.
• list_notifications shows page mentions, comments and replies: a row has page (KEY/N) instead of issue, and page_titles maps those to titles.
• set_subscription follows or unfollows a page: give project and page instead of key (exactly one target).

Write a PRD, then create the tasks

Ask:

Write a PRD page for epic SPD-12, then create tasks from it.

1. get_issue_details for SPD-12 (and get_project_topology for its stories) to learn what the epic is.
2. list_pages with a query to find a parent such as "Product", so the PRD is not a duplicate.
3. create_page with a title, that parent and the PRD as markdown. Writing the key SPD-12 in the text is enough to link the page to the epic.
4. create_issue per work item, each with an idempotency_key; put the page link [PRD](/p/SPD/pages/7) in the description so the task points back at the page.
5. update_page to add the new task keys to the PRD, with the base_rev of a fresh get_page.

References and backlinks

An issue key such as SPD-12 in a page (or in an issue description) links to that issue; in the web app it shows as a chip with the issue's current status. A Markdown link to /p/KEY/pages/N links to a page; the editor's /page command writes one. Sprintree reads these links each time the text is saved, so there is nothing to register: the issue lists the page under Mentioned in pages, and the page lists what links to it under Linked from. Agents read the same lists with get_issue_details and get_page. A viewer sees only the links from places they may read.

Conflicts and undo

• update_page replaces the whole body and needs the base_rev from the get_page you edited. Send the body exactly as get_page returned it, with your changes.
• If the page changed since (a person typing, another agent), nothing is written. The answer shows the current revision and what changed; merge your text with it and retry with that revision. A title-only change needs no base_rev.
• Every body change is a saved version: get_page with include: ["versions"], then version for the old text, and update_page with it to go back.
• Give create_page an idempotency_key and retry safely after an error: the same key with the same arguments within 24 hours returns the first page, with other arguments it is refused.

Use with sysdraw

sysdraw has its own MCP server. Connect both and an agent can move between them. There is no gateway: each server signs you in on its own.

claude mcp add --transport http sprintree https://sprintree.com/mcp
claude mcp add --transport http sysdraw https://sysdraw.app/mcp

Then sign in to each one in /mcp.

Draw an epic

Ask:

Draw epic SPD-40 in sysdraw and put the diagram in its description.

1. Sprintree get_project_topology with {"project":"SPD","parent":"SPD-40","depth":"all"}: the epic's stories and tasks and their links.
2. The agent writes a Mermaid flowchart: one node per story, one arrow per blocks link (or per parent → child when there are no links).
3. sysdraw create_drawing with share: "view" and idempotency_key: "sprintree:SPD-40:v1" (a retry returns the same drawing). The answer's share.markdown is the image line.
4. Sprintree update_description with the base_rev from get_issue_details, adding share.markdown under ## Diagram; or add_comment to leave the description as it is.

Sprintree shows a diagram only for a sysdraw view link that anyone can open, written as an image line (![title](link)); a bare link in Markdown stays a link. When the text is saved, Sprintree asks sysdraw about each new link. A comment or edit link, a link for signed-in sysdraw users only, or a link that is no longer shared is refused with the reason. If sysdraw cannot be reached, the call fails and can be retried.

The link is public. Anyone who has it can see the diagram and everything written in it, issue titles included, without signing in, even if the project is private. Only draw what you are happy to share outside the project. Removing the diagram from the issue does not close the link: turn it off in sysdraw.

Tasks from a diagram

Ask:

Turn sysdraw drawing aB3dE5fG7hJ9 into tasks under SPD-41.

1. sysdraw get_drawing with detail: "full": nodes (handle, label, type) and edges (from, to).
2. Sprintree create_issue per node, with an idempotency_key built from the node handle, so a retry creates nothing twice.
3. link_issues with kind: "blocks" per edge.
4. Optionally sysdraw comment on the drawing with the new keys.

Undo

• Sprintree description edits are versioned: get_issue_history, get_description_version, then update_description with the old text (a diagram removed since is checked with sysdraw again).
• unlink_issues removes a link; an issue made by mistake is removed with delete_issue (its children lose their parent).
• In sysdraw, list_versions and restore_version undo a drawing change. A restore is refused if someone edited the drawing since, so ask before forcing it.

The same flows from sysdraw's side are at sysdraw.app/docs/mcp#sprintree.

Mermaid diagrams

No extra tool: an agent writes a fenced code block with the language mermaid in any Markdown it sends (a description, a comment, a page) and Sprintree shows it as a diagram, with a Show source toggle. The diagram is drawn in the browser; the text stays in the document, so reading it back with get_issue_details or get_page returns the same block unchanged, and it can be edited and compared like any other text.

```mermaid
flowchart LR
  SPD-41 --> SPD-42
```

• The language must be exactly mermaid; other code blocks stay code.
• A diagram of up to about 8,000 characters or 150 lines (a mindmap: 80 lines) is drawn when it comes into view. A bigger one shows its source and a Large diagram - Draw button; over about 20,000 characters or 500 lines (a mindmap: 300 lines) it is never drawn. A diagram with an error shows its source and the parser's message.
• Not supported, the source is shown instead: config directives (%%{init: …}%% lines), front matter (a leading --- block, a title included) and image shapes. Clicks and links in a diagram do nothing. An issue key written inside the block is not a reference to that issue.

Live events in Claude Code

With the plugin, Sprintree events reach a running Claude Code session as they happen, one line each: a mention, an assignment, a reply, a review request, a status change on the project, and /claude … comments you write yourself (a command for your agent), which reach the session only when you turn commands on (see below). There is no polling and no daemon: the plugin's sprintree-events monitor holds one connection open.

Set up

1. Install the plugin: claude plugin marketplace add SNebiyev/sprintree-agent && claude plugin install sprintree@sprintree. If it is already installed, update it instead: claude plugin marketplace update sprintree && claude plugin update sprintree@sprintree. Without the update the monitors never arrive.
2. Run /sprintree:connect. It opens a page on /connect: check that the code matches the one in your terminal, then approve. There is nothing to paste.
3. When the plugin signs in to Sprintree's MCP, tick Account settings on the consent page. set_stream_focus and claim_stream_event need the mcp:account scope, which is off by default. Without it a session does not act on /claude comments and says why. Already signed in without it? Run /mcp in Claude Code, pick sprintree and sign in again with Account settings ticked.
4. Restart the Claude Code session: monitors start with it.
5. Say which project the session follows: put {"project": "KEY"} in .sprintree.json in the repository, or set SPRINTREE_PROJECT. With neither, the monitor stays quiet.

Turn on /claude commands

Since plugin 0.6.0, /claude … comments reach a session only if you turn commands on. Plugin 0.7.0 adds /sprintree:commands for that (update the plugin as the list above says). Type it in the Claude Code terminal:

• /sprintree:commands on (or off) switches commands for this session only, with no restart. The monitor reconnects by itself and prints sprintree-events /claude commands on (or off).
• /sprintree:commands default on (or off) is the setting for every session that has none of its own, running ones too: they reconnect within seconds.
• /sprintree:commands status shows what applies, where it comes from, and whether the server confirmed it.
• The older way still works: add {"env": {"SPRINTREE_COMMANDS": "on"}} to your own ~/.claude/settings.json, merged into what is already there, and restart the session. Put it in your user settings, not in a repository's committed .claude/settings.json.

A session uses its own setting first, then the default, then SPRINTREE_COMMANDS, then off. The switch is only in the terminal because anyone who gets into your Sprintree account could otherwise give work to your agent: no MCP tool or web button turns it on. A command goes only to sessions with commands on. A session it was addressed to that has them off prints sprintree-events /claude command on KEY-12 skipped · /claude commands are off in this session instead. A plugin older than 0.7.0 counts as on for routing, so if its SPRINTREE_COMMANDS is off, a command addressed to it is dropped without a skipped line: update the plugin. Every other line (mentions, assignments, replies, review requests, status changes) arrives either way. While commands are off, the monitor's connected line ends with /claude commands off. Live connections in Settings shows a /claude on or /claude off badge per session (none for a plugin older than 0.7.0).

Several sessions on one project

Each session has a label, shown under Live connections in Settings. It is the git branch (the repository folder's name when HEAD is detached), lower-cased, with every character other than a-z, 0-9, dot, underscore and hyphen (a / too) turned into a hyphen, cut to 40 characters. So Feature/Login is feature-login. When two sessions share a name, each row in Live connections adds the first 4 characters of its session id, the same as session= in the terminal's connected line. Events reach the sessions like this:

• Mentions, assignments, replies, review requests and status changes reach every session of the project.
• A /claude … comment goes to the session that has that issue in focus (set_stream_focus). With no session focused on it, it goes to every session with commands on, and the first to call claim_stream_event answers while the others stay silent.
• /claude @<name> … (a space after /claude) goes to the sessions whose label is that name, ignoring case. Type the label as Live connections shows it: @feature-login, not @Feature/Login, which matches no label. If two sessions share the label, both get it and the first to claim answers. If no live session has that label, it goes to every session with commands on, as if there were no name.
• A session it is addressed to (the issue is in its focus, or @label names it) that has commands off prints a skipped line, and the command goes to no other session.

Safety

• The stream carries keys, usernames, ids, counts and session labels only (a /claude command adds the @name it was addressed to), never titles, descriptions or comment text. The agent reads the content over MCP, with your rights.
• The stream token (itns_…) opens nothing but the stream. It is created only by approving in the browser, never through MCP, so it never appears in an agent transcript.
• Revoke it in Settings → API & MCP under Personal access tokens, or with revoke_stream_token. The Live connections list on the same page shows each session and detaches it.
• Changes your agent makes through MCP never wake a session.

Limits

• It needs a Claude Code version with plugin monitors, in an interactive session. Plugin monitors are not available with Bedrock, Vertex or Foundry.
• At most 10 live connections per account. An 11th is refused until one is detached or has gone quiet.

Tools

These tools need the mcp:account scope (an OAuth grant, not a token): list_stream_tokens, revoke_stream_token, list_stream_sessions, detach_stream_session, set_stream_focus, claim_stream_event.

Read tools

Need the mcp:read scope. They never change anything.

• list_projects

  List your projects: those you are a member of plus featured ones (featured: true; read-only).

• get_project_topology

  List a project's issue tree (epics → stories → tasks), level by level or a whole subtree.

• search_issues

  Find issues by text and filters…

• get_issue_details

  Get one issue: metadata (status, priority, assignee, reporter, parent, due, created, updated, labels, subscribed), base_rev…

• list_notifications

  List your Sprintree inbox, newest first: mentions, assignments, comments and replies on issues you follow, status changes, edits of issues you follow…

• list_statuses

  List a project's statuses (board columns) in board order with their key, name and category (todo, in_progress, review, done).

• list_events

  List events, what changed in the projects you are a member of (issues created, moved or edited, comments, members, files), oldest first; pass project to read…

• list_project_activity

  Read the activity feed of a project (or, without project, of every project you are a member of), NEWEST first: issues created, deleted, retitled, moved…

• get_event

  Get one event by the id list_events gives, with its full payload (still no content: use get_issue_details for that).

• list_labels

  List a project's labels with their colours (needs read access).

• list_members

  List a project's members (owner, editors, viewers; by name) with username, display name, role and last active (members only).

• get_issue_history

  Show the change history (audit trail) of an issue, newest first and all of it, page by page (no entry is out of reach): what changed (field, from → to), who…

• get_description_version

  Read an earlier (or the current) version of an issue's description as Markdown.

• list_issue_files

  List the files attached to an issue: id, name, size, type, who uploaded it and when, and the download URL (open it with the same token).

• get_issue_stats

  Count issues, per status, assignee or type, in one project or across all yours.

• list_invitations

  List pending project invitations with their ids.

• get_comment_versions

  Show the earlier bodies of an edited comment, newest first with the current one on top (comment ids are shown by get_issue_details and list_notifications).

• get_preferences

  Show your settings: time zone (which decides what "today" and "overdue" mean and when the daily digest arrives), e-mail notification mode…

• list_connected_accountsneeds the account scope

  List the Google, GitHub and GitLab accounts that can sign in as you (the "Connected accounts" of Settings > Security): provider, the e-mail address the…

• list_mcp_changes

  List what changed in this MCP server, newest first: tools that were added and changes to existing tools (new arguments, new output), each with its date.

• list_project_repositories

  List the GitHub repositories connected to a project (any member or a viewer of a public project may read them): id (what disconnect_repository takes), full…

• list_issue_commits

  List the commits that name an issue key in their message (any member or a viewer of a public project may read them), newest first: repository, short SHA…

• get_slack_notifications

  Read where a project posts to Slack (members only): workspace, channel, state (pending: no channel yet; active; error: see last_error), the events…

• list_criteria_templates

  List the criteria templates of a project, or of one epic (epic: its key), both kinds, in the order issues get them (project owner only).

• get_work_context

  Everything needed to start work on one issue in a single bounded call (read-only): its fields, the project description (the team's notes), the parent chain up…

• list_views

  List the saved views of a project: your personal views first, then the project's shared ones, each by name.

• list_cycles

  List a project's cycles (numbered time boxes for issues), newest first: dates, state, goal, open/done counts, rev.

• get_run

  Without run: your latest runs, open ones first.

• list_stream_tokensneeds the account scope

  List your live-events stream tokens.

• list_stream_sessionsneeds the account scope

  List your stream sessions (live ones and those seen in the last 10 minutes).

• list_pages

  List a project's page tree.

• get_page

  Get a page: title, parent, breadcrumbs, url, base_rev (update_page takes it), version, who/when, and the body as the exact Markdown update_page expects…

Write tools

Need the mcp:write scope. The few tools marked needs the account scope (changing your preferences, answering invitations) also need mcp:account, which mcp:write does not include: only an OAuth grant can hold it, you tick Account settings on the consent screen (it is off by default), and a personal access token never has it.

• create_issue

  Create an epic, story or task.

• update_issue_status

  Move an issue to another status of its project (needs edit access).

• update_issuecan overwrite data

  Edit an issue: change its title, priority, assignee, due date, parent, cycle, type or labels.

• update_descriptioncan overwrite data

  Replace an issue's description with Markdown.

• add_comment

  Comment on an issue (key) or on a page (project and page number) in Markdown.

• resolve_thread

  Mark a comment thread as resolved (its highlight in the description is hidden).

• reopen_thread

  Reopen a resolved comment thread.

• mark_notifications_read

  Mark inbox notifications as read: pass ids (notification ids from list_notifications) or all=true for every unread notification.

• snooze_notifications

  Hide inbox notifications until a later time: they leave the inbox and the unread count, then come back unread at the top.

• mark_notifications_unread

  Mark read inbox notifications as unread again, by ids from list_notifications.

• archive_notifications

  Mark inbox notifications Done: they are marked read and leave the inbox…

• restore_notifications

  Bring snoozed or Done notifications back to the inbox at their original place (the undo of snooze_notifications and archive_notifications); unread=true also…

• create_project

  Create a project owned by you.

• example_issuescan overwrite data

  Count, add or remove template example issues.

• update_projectcan overwrite data

  Change a project's name, description, key (issue prefix), colour or visibility.

• create_label

  Create a label in a project (needs edit access).

• attach_file

  Upload a file to an issue (needs edit access).

• create_status

  Add a status to a project's board (needs edit access), at the end.

• update_status

  Rename a status, change its colour or (for a custom status) its category (needs edit access).

• reorder_statuses

  Put a project's statuses in a new board order (needs edit access).

• delete_statuscan overwrite data

  Delete a custom status (needs edit access).

• edit_commentcan overwrite data

  Replace the text of one of your own comments with Markdown.

• delete_commentcan overwrite data

  Delete a comment.

• delete_issuecan overwrite data

  Delete an issue.

• update_labelcan overwrite data

  Rename a label and/or change its colour (#rrggbb).

• delete_labelcan overwrite data

  Delete a label from a project and remove it from every issue that carries it.

• delete_filecan overwrite data

  Permanently delete an uploaded file (an attachment or an image in a description) by its id, the uuid in its /api/files/<id> URL.

• bulk_update_issuescan overwrite data

  Change the same fields on up to 200 issues at once, of one project or of SEVERAL…

• move_issue

  Move an issue's card to another board column (status) and/or reorder it among the other cards, like dragging it (needs edit access).

• archive_projectcan overwrite data

  Archive a project.

• invite_membercan overwrite data

  Invite someone to a project by e-mail (project owner only).

• change_member_rolecan overwrite data

  Change a project member's role to editor or viewer (project owner only).

• remove_membercan overwrite data

  Remove a member from a project (project owner only), or pass your own username to leave it (the owner cannot).

• cancel_invitationcan overwrite data

  Cancel a pending invitation (project owner only).

• set_subscription

  Follow or stop following an issue (key) or a page (project and page number).

• set_watch

  Set how you follow a whole project (members only, a non-member who can read a public project is refused): all (every issue, as if subscribed to each)…

• delete_archived_projectcan overwrite data

  Delete an archived project you own, or only its files.

• answer_invitationneeds the account scopecan overwrite data

  Answer an invitation addressed to you (ids from list_invitations): accept=true joins the project with the invited role, accept=false declines it.

• set_preferencesneeds the account scopecan overwrite data

  Change your settings (time zone, e-mail, theme, formats, display options); only the fields you pass change.

• pin_project

  Pin or unpin a project (top of your sidebar and list_projects) and/or set its unseen badge (members).

• move_project

  Move a project inside its section and group of your sidebar (pinned, your projects, or public ones): after or before another project there, or to the top or…

• link_issues

  Link two issues of the same project (needs edit access).

• unlink_issuescan overwrite data

  Remove a link between two issues (needs edit access).

• set_reviewerscan overwrite data

  Ask members to review an issue (add) or take reviewers off (remove): usernames or 'me'.

• submit_review

  Answer a review request as its reviewer (404 if not asked).

• disconnect_repositorycan overwrite data

  Disconnect a GitHub repository from a project (needs edit access; the repository id is from list_project_repositories).

• set_slack_notifications

  Change a connected project's Slack notifications (needs edit access; connect in the web app first).

• set_criteriacan overwrite data

  Change the acceptance criteria (ac) and definition of done (dod) checklists of an issue (needs edit access; a list the project switched off refuses writes).

• check_criteria

  Tick or untick items of an issue's acceptance criteria / definition of done (needs edit access).

• set_criteria_templatescan overwrite data

  Replace the criteria templates of one list (kind ac or dod) of a project, or of one epic (epic: its key): the items an issue gets a copy of when it is created…

• save_view

  Create, change or copy a saved view.

• delete_viewcan overwrite data

  Delete a saved view by its id (list_views): your own personal views, or the project's views with edit access.

• save_cyclecan overwrite data

  Create, change, start, complete or delete a cycle (one is active per project).

• set_commitments

  Commit someone to issues for a week (remove=true: take them off).

• save_project_groupcan overwrite data

  Create, change or delete one of your project groups: the folders of your own sidebar (only you see them; no effect on access).

• start_run

  Open your agent run for a label (the repository or task), or get the open one: calling it again answers the same run.

• update_lanecan overwrite data

  Create or change one subagent's lane of an open run: type, phase, status (running, done, failed, blocked), a short checkpoint…

• close_run

  Close a run when the session's work is over; start_run then opens a new one for the label.

• revoke_stream_tokenneeds the account scopecan overwrite data

  Revoke a stream token (id from list_stream_tokens) at once; no undo.

• claim_stream_eventneeds the account scope

  Take a command (ev of a stream frame) first-come.

• set_stream_focusneeds the account scope

  Say which issue a stream session works on, so /claude comments about it reach it; empty issue clears.

• detach_stream_sessionneeds the account scopecan overwrite data

  Close a stream session; its id cannot reconnect for 10 minutes.

• join_projectneeds the account scope

  Join a public project as a viewer so it is listed in list_projects and your sidebar and you can watch it.

• create_page

  Create a page (edit access); without parent it is top level, without after it goes last.

• update_pagecan overwrite data

  Change a page's title and/or whole body (edit access).

• move_page

  Move a page with its subtree under another parent and/or reorder it (edit access).

• delete_pagecan overwrite data

  Delete a page and its subpages (edit access): page.