MCP for Everyone, Cache Metrics & Job Notifications
More Transparency, Fewer Barriers
In recent updates we introduced Skills and MCP Servers so you can extend your AI agents with external tools and reusable instructions. Now we're taking it further: MCP is available on all plans, runner jobs show detailed cache metrics, and you'll be notified automatically when a job finishes or needs your attention.
MCP Integration: Now on All Plans
MCP integration was previously limited to the Pro plan. That changes now: all users -- including those on the Starter plan -- can configure MCP servers and give their agents access to external tools.
In practice this means you can connect your agents to GitHub, Slack, databases, or any other MCP server -- no upgrade required. The REST API remains a Pro feature, but agent-side tool integration is now open to everyone.
Why This Matters
MCP servers are the easiest way to extend what your agents can do. A Fetch MCP lets agents read external documentation. A GitHub MCP gives them access to issues and PRs. Making this available to every team -- regardless of plan -- significantly lowers the barrier to entry.
If you haven't set up MCP servers yet, our previous post on Skills & MCP Servers has a step-by-step guide.
Cache Token Metrics in Job Views
Runner jobs now display detailed cache information. For every job you can see:
- Cache Read -- how many tokens were served from the prompt cache
- Cache Write -- how many tokens were written to the cache
- Cache Hit Rate -- the percentage of input tokens served from cache
These values appear in the ticket runs panel, the job list, the active runners overview, and the streaming job view.
What This Gives You
A high cache hit rate means your runner is reusing recurring context from cache instead of reprocessing it. That saves tokens, reduces latency, and lowers cost. A low rate may indicate that your prompts vary significantly or that the cache expires between jobs.
Previously these numbers were buried in runner logs. Now you see them at a glance -- right in the UI.
Notifications for Runner Jobs
Until now you had to check manually whether a runner job was done. That's over. Spedy now sends automatic notifications when:
- Job completed -- the runner finished the job successfully
- Job failed -- something went wrong
- Job waiting for input -- the agent needs your response to continue
Notifications go to the ticket assignee and all watchers. In multi-stage pipelines, only the final stage triggers a notification -- intermediate steps don't create noise.
The new notification types appear in your inbox with dedicated icons and deep links straight to the ticket. You can configure them per project or globally, just like any other notification type, and optionally receive them via email.
Run Runners with a Claude Max Subscription
A change for teams hosting their own runner: the Anthropic API key is now optional. If you have a Claude Max subscription, the runner can use Claude CLI's built-in OAuth authentication -- no separate API key needed.
The setup script automatically detects whether Claude CLI credentials exist at ~/.claude and mounts them into the runner container. For teams that already have a Max subscription, this eliminates the need for a separate API key.
Smaller Improvements
- Budget alert at top of dashboard -- The budget warning now appears at the very top of the dashboard instead of being buried between other sections. You'll see budget issues immediately when opening the dashboard.
- Optimized time tracking view -- The time tracking pages (ticket time entries, time overview) have been reworked for better clarity and performance.
Summary
| Change | What it means |
|---|---|
| MCP on all plans | Every team can use MCP servers -- no Pro plan required |
| Cache metrics | See at a glance how efficiently your runner caches |
| Job notifications | Get notified when jobs finish or need input |
| Claude Max support | Run runners without a separate API key |
| Budget alert on top | Budget warnings immediately visible |