SprintFlint

Opinionated sprint tracker. Read/update tickets, sprints, velocity from Claude/Cursor/Zed.

Live · Protocol 2025-06-18 · Streamable HTTP

SprintFlint MCP Server.

Sprint state in every editor.

Public Model Context Protocol endpoint at mcp.sprintflint.com. Drop one JSON block into Claude Desktop, Cursor, or Zed (or one CLI command in Claude Code) and the agent reads your sprint, files issues, and posts comments without leaving the chat.

Generate configGet a token

Endpoint

One URL, one auth header. Same bearer token as the REST API.

POST https://mcp.sprintflint.com
Authorization: Bearer YOUR_TOKEN
Content-Type: application/json
Accept: application/json, text/event-stream

Generate a token from your account settings. Free tier covers 300 tickets; no card.

Tools

Day-one tools, mapped 1:1 onto the actions a developer takes during a sprint.

list-projects

Search projects in your organisation (paginated). Returns id, name, slug, and counts.

list-sprints

Search sprints inside a project (paginated). Returns sprint id, goal, start/end dates, and issue counts by status.

list-issues

Search issues inside a sprint (paginated). Filter by status. Returns id, title, status, story points, assignee, labels.

my-issues

Issues assigned to the calling user. The default opening question for any agent: "what am I working on?"

get-issue

Read a single issue by reference (e.g. PROJ-123) including full markdown description and comment thread.

create-issue

File a new issue into a sprint. Title, description, sprint id, optional priority/labels/assignee.

update-issue

Update title, description, status, story points, or assignee on an existing issue. All mutations reversible from the UI.

create-comment

Post a markdown comment to an issue. Used by agents to log PR links, status updates, and surfaced blockers.

What's deliberately not exposed

Two operations we did not ship on day one; both irreversible (or hard to undo) and unsafe to expose blindly to a model that occasionally hallucinates an argument. Update operations are available (every mutation is reversible from the UI) but destruction stays a human decision.

  • delete-issue: silent destruction of work history. Cost asymmetry too high. Click delete in the UI.
  • delete-sprint: silent breakage of velocity history and sprint planning. Use the board.

Read the design rationale: Sprint state in your agent: the practitioner's guide.

Connect in 30 seconds

Pick your editor. The MCP Config Generator emits the right snippet for each.

Claude Desktop

Paste mcpServers block into claude_desktop_config.json. Restart.

Setup

Cursor

Settings → MCP → Add server. Same JSON shape.

Setup

Claude Code

One CLI command: claude mcp add sprintflint --transport http ...

Setup

Zed

context_servers block in ~/.config/zed/settings.json.

Setup

Auth, scoping, security

  • Auth: Authorization: Bearer YOUR_TOKEN. Same token as the REST API.
  • Scope: tokens are scoped to your organisation. No cross-org access.
  • Revocation: tokens listed by name + last-used time in account settings. Revoke instantly.
  • Recommendation: one token per device (e.g. claude-desktop-macbook, cursor-work-laptop). Easier to spot orphans.
  • TLS: Heroku ACM with Let's Encrypt. HSTS preloaded.
  • Rate limit: 600 requests/min/token. 429 with Retry-After header on exceed.

FAQ

Everything you need to know about the MCP server.

What protocol version does the server speak?

2025-06-18 (current stable). Also accepts 2024-11-05 for older clients.

Is the MCP server free?

Yes; same plan rules as the REST API. Free tier (300 tickets) gets full MCP access. Paid plans (£5/user/month) get unlimited.

Does the server support resources or prompts (the other two MCP primitives)?

Tools today. Resources and prompts are on the roadmap; see /roadmap. The capabilities handshake currently advertises tools, prompts, resources, logging; tool calls are the only fully-implemented primitive on day one.

Can I use this server with my own agents (LangChain, autogen, etc.)?

Anything that speaks MCP works. Bring your own client. Some non-MCP clients work better with the REST API: same data, same auth.

What happens if the agent hallucinates a sprint or ticket id?

The server returns a structured error. The agent sees the error and (usually) corrects itself. We deliberately omitted destructive tools so the worst-case is a failed read, not a silent delete.

Is the server hosted, or can I self-host?

Hosted only for now (mcp.sprintflint.com). Self-hosting MCP servers makes sense for offline-first tools, less so for SaaS; your sprint data already lives on our servers.

How do I report bugs or request tools?

Email [email protected]. We read everything.

Still have a question? Email us; we usually reply within an hour.

Plug your agent in

Free up to 300 tickets, forever. No credit card. Same bearer token works for MCP, REST, and OpenAPI.

Get my configRead the protocol primer

We use cookies to provide essential functionality and improve your experience. Analytics cookies help us understand how you use SprintFlint.

Accept All Essential Only

Related Servers