MCP on Every Plan – Connect AI Tools Directly to Spedy

Note (2026-09-02): The coding runners, live previews, agent skills and runner settings described in this post have been removed from Spedy. Spedy no longer runs agents — your own agent (Claude Code, Codex, Cursor, …) works with Spedy directly via MCP. Background in Runners, live previews and agent skills removed. Everything else described here is still available.

You can now connect MCP-capable AI tools like Claude Desktop or Cursor to Spedy — included on every plan. We've also upgraded runner jobs with new metrics, smarter notifications, and better control when agents need your input.

---

MCP on All Plans

MCP (Model Context Protocol) was previously a Pro-only feature. That changes now: MCP is included on every plan.

What you can do with it:

• Search projects directly from Claude Desktop or Cursor
• Read and comment on tickets without opening a browser
• Build AI workflows that use Spedy as a data source

The REST API remains a Pro feature. MCP is the recommended path when you want to connect AI tools to Spedy.

---

Runner Jobs: More Visibility

Runner jobs now show cache token metrics directly in the overview: how many tokens were read from cache, how many were written, and the hit rate as a percentage. You'll see this in the ticket detail view, the Jobs tab, the dashboard, and during live execution.

At a glance, you can assess how efficiently a runner is operating — useful for understanding costs and improving prompts.

---

When the Agent Has a Question

If the orchestrator detects a blocker during the planning phase — missing information, for example — the streaming view now shows an amber "Answers required" panel. The approve button disappears, and you get an input field to answer the agent's questions directly.

The AI suggests, you decide. The pipeline only continues when you give the go-ahead.

---

Run Runners Without a Separate API Key

If you use Claude Max, you no longer need a dedicated Anthropic API key for runners. The supervisor automatically detects whether a key is configured. If not, it uses local OAuth authentication via ~/.claude.

To set it up:

1. Make sure you've run claude locally at least once
2. Leave the Anthropic API Key field empty during setup
3. The supervisor automatically mounts ~/.claude into the runner containers

Especially convenient for developers who already use Claude Max and don't want to manage a second key.