Your agent, with Spedy as its context.
Spedy runs no agents: you bring your own, Claude Code, Codex, Cursor or any other MCP client, with your own key. Through Spedy’s MCP server it fetches ticket, wiki and knowledge base, works in your environment and reports back to the ticket.
- tickets_claim NORD-214 · 30 minok
- work-context_get NORD-214ok
- wiki_search "caching strategy"ok
- timers_start NORD-214draft
- tickets_report · PR #143running
One command, and Spedy is in your agent.
Under MCP in your account there are ready-made snippets for Claude Code, Cursor and Codex, with the address of your subdomain. About 60 tools for tickets, to-dos, wiki, timers, knowledge base, pull requests and pipelines.
- With the user’s permissions
- Every tool checks role, permissions and project access. A read-only grant removes every writing tool.
- Tokens with limits
- Per token: only certain projects, read-only, no deleting, an expiry date. Only a hash is stored.
- claude mcp add --transport http spedy …Claude Code
- "mcpServers": { "spedy": … }Cursor
- [mcp_servers.spedy]Codex
Agents with their own identity.
You create agents as their own users, each owned by a person, with no login, only tokens. They appear like team members: as assignees, in comments, in the history.
- Not a seat, but a limit
- Agents do not count as licensed seats. Up to 25 agents per organization.
- Rules for agents
- Chosen status moves need human approval; the agent then only files a suggestion. Agent PRs can be flagged as requiring review.
- Weekly cap and time factor
- Per agent, a weekly cap on booked hours and its own time factor.
Claim, work, report.
Before working, an agent can claim the ticket: an exclusive reservation with an expiry. At the end, one call writes a report (result, changes, open points, PR), stops its timer and releases the ticket.
- Delivery trace per ticket
- Claims, reports, status moves, branch, PR with reviews and merge, pipelines and booked time in one timeline. What is not tracked is listed as a gap.
- Time as a draft
- What the agent books waits for your approval before it reaches the budget, the client or MOCO.
- Claimed by claude-nordwind09:12
- PR #143 opened10:41
- Report: result · changes · open10:42
- Review by Lea Brandt14:05
- Time 1.5 h approved14:06
What it does not do
So there are no surprises in the demo.
- Spedy does not run, host or pay for agents. There is no built-in coding agent, no runners, no preview environments.
- No AI features inside Spedy itself: no chat, no ticket writing, no triage. The only ML is self-hosted embeddings for semantic search.
- Spedy is an MCP server, not an MCP client: you do not connect third-party MCP servers inside Spedy.
Frequently asked questions
- Does our code leave the building?
- The agent runs on your side; Spedy executes no code. Via MCP the agent fetches tickets, wiki and knowledge; what lands in Spedy is what it writes: comments, status moves, time drafts. The Git connection delivers commit messages and PR and pipeline status, not source code.
- Which agents work?
- Any MCP client that speaks HTTP. For Claude Code, Cursor and Codex there are ready-made snippets in the app. From the ticket menu, Spedy also builds a prompt from the ticket and opens it in the tool you pick: Claude Code in the browser, desktop tools such as Cursor, Windsurf or Zed via deeplink where the tool supports it. Internal comments are left out. No SSE streaming.
- Can an agent delete things?
- Only if its token allows it. For agent tokens, “no deleting” is the default.
- Do we need a separate agent user?
- No. An agent can work with your own account. Separate agent users make visible what the agent did and allow their own rules and limits.
Let us talk about how you work.
Spedy is only available through a demo call. Request a demo or book a slot directly.