Sevalla Changelog #0099 Claude, MCP, and Production Control
Sevalla adds 27 MCP tools for Claude, plus a refreshed dashboard, docs and API reference, maintenance mode, egress IPs, and Redis and Valkey support.
Establishing connection…
Signal detected.
Link established.
Baud here with a 56k Sevalla changelog transmission.
Seven platform updates shipped, led by a major upgrade to Sevalla’s MCP server, a new listing in Claude’s connector marketplace, and a refreshed dashboard, docs, and API reference built on one shared design. Claude can now work directly with your Sevalla account through 27 dedicated tools for deploying, debugging, configuring, and managing production services.
The rest of this transmission brings the same idea to more of the platform: clearer workflows, more control over production, and less infrastructure and tooling to manage.
A cleaner dashboard, docs, and API reference without relearning Sevalla#
Sevalla’s dashboard, documentation, and API reference now share one design system.
The important part is what did not change: the sidebar, service tabs, and everyday workflows are still where you expect them. This is not a redesign that makes experienced users relearn the product. It is a consistency upgrade across the surfaces developers use to deploy, operate, troubleshoot, and automate production services.
In the dashboard, list and service pages now follow the same structure, settings use consistent sections, and irreversible or service-disrupting actions live in a predictable Danger zone. Statuses include a color, icon, and label instead of relying on color alone.
Navigation is also faster. Breadcrumbs now appear across the dashboard, and ⌘K on macOS or Ctrl+K on Windows and Linux lets you jump directly to pages, services, and projects. Long names, hostnames, and commit messages stay accessible instead of being silently truncated, and every screen now works down to phone width.
The dashboard is checked against WCAG 2.2 AA in both light and dark themes.
The docs received the same treatment. Navigation now follows the dashboard sidebar, labels match the interface, screenshots follow your selected theme, and old links redirect to the corresponding new pages. Each page now opens with a clear description and includes metadata written for search and AI agents consuming /llms.txt.
The API reference now uses the same design and product organization as the docs. Guides for errors, rate limits, and pagination have been rewritten, and a new long-running operations guide explains which operations return before the underlying work completes and how to poll them.
The CLI reference now covers every command in CLI 1.10.1, and the API reference documents v3 only, with old v2 links redirecting to the matching v3 pages.
Nothing changes in how your services, API, CLI, or MCP server work. You get a more consistent platform surface without a migration or workflow reset.
Give your AI agent real platform tools, not a code sandbox#
The Sevalla MCP server now has 27 dedicated tools for the workflows developers use every day.
Instead of generating code, running it in a sandbox, and hoping the result maps cleanly to the platform, agents can work directly with Sevalla operations.
They can:
- Discover applications, databases, and static sites
- Trigger deployments and rollbacks
- Read deployment, runtime, and access logs
- Fetch metrics
- Manage environment variables
- Scale processes
- Add custom domains
- Create applications, static sites, and databases
For less common operations, generic sevalla_api_* tools can find and execute operations across the Sevalla API.
That matters when AI moves beyond answering questions and starts touching production. The difference between a generic execution environment and explicit platform operations is control.
Sevalla is also now available in the Claude connector marketplace, giving Claude users a direct path to connect their Sevalla account without manually configuring the MCP server.
Once connected through OAuth, Claude can use the same Sevalla tools to deploy and roll back applications, investigate failed deployments, inspect logs and metrics, manage configuration, create services, and search the Sevalla documentation. The connection uses a Sevalla API key created during authorization, so Claude operates within that key’s permissions and can be disconnected by revoking it from the Sevalla dashboard.
Every MCP tool is marked as read-only or destructive. Compatible clients such as Claude can run read-only operations directly and ask for confirmation before making destructive changes. The server never returns environment variable values to the agent.
The server also supports MCP protocol version 2026-07-28 while remaining compatible with clients using earlier versions. Existing connections continue working without reconfiguration.
This makes AI-assisted operations much more direct: inspect an application, check a failed deployment, search logs, change configuration, and redeploy without building another layer of scripts around the platform.
See the full MCP tool list.
Move external database and TCP connections without downtime#
Database external connections and application TCP proxies now use new hostnames and ports.
The migration does not require a cutover window. Legacy and new endpoints run in parallel for existing services, so you can update clients on your schedule instead of coordinating an avoidable outage.
For databases, the new endpoint appears under:
Networking > Public networking > External connection
and
Overview > External connection
For application TCP proxies:
Networking > Public networking > TCP proxies
If a service still has a legacy endpoint, Sevalla shows the migration deadline and lets you reveal the old connection details so you know exactly what needs replacing.
Internal connections between Sevalla services are unaffected.
This is infrastructure maintenance, not a migration project for your infrastructure team.
Read the database networking docs or application networking docs.
Know exactly where application traffic leaves Sevalla#
Some external services will not accept traffic until you allowlist their source IP.
You can now see the public addresses used by outbound traffic from any Sevalla application under:
Networking > Egress IP addresses
This gives production teams the information they need to connect Sevalla applications to databases, APIs, firewalls, and other systems with IP-based restrictions without guessing where traffic originates.
Applications running in Oracle Cloud data centers receive a single fixed IP address.
Applications running in Google Cloud data centers show the IPv4 ranges Google publishes for the selected region, with a Copy all option for the complete list.
The data is also available through the API using Get egress IPs.
Read more in the Networking docs.
Preview branches in parallel, not in a queue#
Pull request previews are most useful when developers can actually use them while work is happening.
Static site previews for different branches now build in parallel. They also build alongside production deployments instead of waiting behind them.
That means one developer's preview no longer blocks another developer's branch from building.
Pushes to the same preview branch still run sequentially. The latest commit remains the final state you preview.
Production deployments are unchanged and continue running one at a time per site.
We also separated preview state from site state. A preview that is building or has failed no longer marks the entire static site as failed or blocks site updates.
Preview environments should help teams review changes faster, not become another deployment queue to manage.
Put an application into maintenance mode without taking it apart#
Planned maintenance should not require rewriting routing rules or shutting down an application.
Applications now have a Maintenance mode under Settings.
Turn it on and requests across every application domain receive a maintenance page with an HTTP 503 response and Retry-After header. The application itself keeps running, and deployments remain available.
That gives you a cleaner way to handle work such as database migrations while keeping the application operational behind the maintenance response.
Sevalla can serve the default maintenance page, or you can provide any HTTPS URL as a custom page. A Sevalla static site works if you want a lightweight custom maintenance experience while continuing to serve it from the application's own domain.
Maintenance mode is available through the API too:
See the Settings docs.
Manage Redis and Valkey without adding another database tool#
Database Studio now supports Redis and Valkey.
That means you can do common key-value database work directly where the database runs, rather than needing redis-cli or another GUI to inspect production data.
You can browse the keyspace using pattern and type filters, switch between flat and namespace views, work across logical databases, and perform bulk TTL or delete operations.
Dedicated editors support:
- Strings
- Hashes
- Lists
- Sets
- Sorted sets
- Streams
- JSON
You can add, rename, edit, and delete keys, manage TTLs with a live countdown, and work with JSON using highlighting, bracket matching, and folding.
Need lower-level access? Database Studio includes a Redis CLI-style console with autocomplete, command history, and inline command references.
For operational troubleshooting, you can inspect memory, connected clients, throughput, keyspace data, searchable INFO output, the slow log, and active client connections.
Loaded keys can also be exported as .csv or .json.
It is one less tool to install, secure, switch to, and maintain just to understand what your production database is doing.
See the Database Studio docs.
Transmission complete#
Seven releases give production teams clearer workflows and more direct control over deployments, networking, databases, maintenance, and AI-assisted operations without pushing infrastructure management back onto developers.
More improvements on the way.
Awaiting next transmission.
Link closed.