Roadmap

This is the release we’re finishing right now, due in September. Most of it is about MCP: one server that answers real questions, and a safer way to sign in to it.

• Reworked MCP We cut the tool list down to three that do the work: describe_schema, query, and fetch. Ask about services, workloads, clusters, GitOps state, Helm releases, pod events, scorecards, repositories, docs, teams, and people, with filters, grouping, and counts, across every connected cluster. Four older tools are gone: three that never returned data, and search_docs, which query covers. The other list and get tools stay.

• Workload diagnosis Ask why a workload is unhealthy and your assistant gets one health summary back: pod events, GitOps sync state, config drift, rollout status, and right-sizing. It digs into detail only where something looks off.

• MCP sign-in Your assistant sends you through your normal company login instead of asking for a token. No long-lived secret sitting in a config file, and revoking someone in your identity provider revokes their MCP access too. Works with Zitadel and Okta, and tokens keep working everywhere.

• Switch identity provider Moving from one sign-in provider to another used to lock people out, so we fixed it. The Users page now shows who’s affected, links people to the accounts they already have, and removes anyone the new provider doesn’t list, with every removal in the audit log.

• HolmesGPT Point HolmesGPT at Shoehorn and it can ask who owns a service, where the runbook lives, and how a workload is doing across your clusters.

• Zombies that are zombies The Zombies page now lists only running workloads that nothing depends on and nothing has touched in 90 days. Cluster components are left out, and each row names the last thing that happened to the workload.

• Fixes We’ve been chasing bugs across teams, search, Forge approvals, and the admin pages.

A small one, also in September: upgrades to the pieces Shoehorn runs on.

• Dependency upgrades Newer Meilisearch, Go, Valkey, Redpanda, and Cerbos. Meilisearch needs one extra upgrade step, which we describe in the Helm chart README. Your search data moves over instead of being rebuilt.

In October we want one page that explains a service, an honest answer to why something never showed up, and a CLI your assistant can use.

• Service Passport Owner, code, docs, dependencies, and where the service runs, on one page, with a link to the source behind every fact. When a source is stale or missing, the page says so instead of quietly leaving it out.

• Agent coverage Your cluster agent reports what it can see and what it can’t, down to which namespaces it watches, which permissions it’s missing, and how fresh each kind of data is. We’ve spent enough time guessing why a workload never showed up in the catalog, and we suspect you have too.

• A CLI your assistant can use shoehorn query, shoehorn fetch, and shoehorn schema return the same answers as MCP, as JSON, under the same permissions. Coding assistants already live in a terminal, and CI jobs and scripts can run the same commands. Sign in with your company login instead of pasting a token.

• Integration:Auth Sign in and sync your directory with Microsoft Entra ID. We talked about this one back in July. October is when it lands.

November is about what changed around a service, on one timeline.

• Change Lens Commits, deployments, GitOps reconciliations, and cluster events in one ordered list per service. Shoehorn marks whether an item is directly linked to a revision or just happened nearby, and it won’t claim one caused the other. We’d rather show you the order of events and let you draw the conclusion than guess on your behalf.

• Unreviewed changes Each change on the timeline shows how it got in: approved by another person, merged by its author alone, approved only by a bot, or pushed straight to the branch. It doesn’t matter who or what wrote the code. You can filter on it in the web app, over MCP, and in the CLI.

• Argo CD apps in the right place When one Argo CD instance deploys to several clusters, each app shows up on the cluster it actually runs on. Apps built from more than one source keep their repository and revision. Neither works today, and the timeline needs both.

December, and the last feature release before we freeze. This one is about where to start your day.

• Needs your attention Ask your assistant what needs your attention and you get a short, ranked list back. Drift, failed rollouts, security findings, scorecard failures, ownership gaps, and unreviewed changes running in production, collapsed into one item per real problem. Unreviewed changes only show up when we can tell which commit is actually running. Each item says why it ranks where it does and links to the evidence, which gives an assistant enough to start working on it. Your home page and each team page show the same list.

• Operations notifications Right-sizing, missing resource requests, and overprovisioned workloads become signals you can subscribe to.

Feature freeze in January. This is the last stretch before 1.0, and we’re not adding anything new to it.

• Hardening We spend this release hunting edge cases and bugs, and making sure upgrades, backups, and restores from older versions land cleanly.

• One visual language Topology, the org chart, and Forge get the same look as the rest of the app.

We aim to release 1.0 in March 2027. That’s ambitious, and we know it, so we’d like your feedback on the way there. The idea behind it hasn’t changed: your tools already know what happened, and Shoehorn shows how it fits together.

• 1.0 The lenses ship as one thing: passport, change, attention, and agent coverage over a single service graph. No hand-written catalog YAML to keep any of it working.

• The same answers over MCP Anything you can see in the web app, your AI assistant can ask for, with the same permissions and the same caveats. A “probably” in the web app stays a “probably” when an assistant reads it. Shoehorn doesn’t run or call any AI model. Your assistant does the reasoning, and we hand it facts with sources.

• Fresh install Connect one source and reach something worth looking at in under five minutes. That’s the bar we’re holding ourselves to.

After 1.0. We’re pencilling in May 2027, and this is the part of the roadmap most likely to move.

• Cost Lens Spend per service, team, environment, and cluster. Kubernetes compute and storage plus UpCloud resources, with a FOCUS import for billing data from anywhere else. Every figure says whether it’s an estimate or a real bill, idle cluster spend gets its own line, and a jump after a deploy shows up next to the deploy.

• GitOps outcome One view of sync state, drift, and the resources each app manages, across all your Argo CD instances and Flux clusters, with the owning team on each. When a revision hasn’t reached an environment, you see the revision you wanted, the one that landed, and what’s blocking the rest. For Flux that includes sources that failed to fetch, dependencies that aren’t ready, controller health, and the images your automation picked. You’ll still sync, reconcile, or roll back with Argo CD or Flux’s own tools, and we link you to the right place.

• Impact code search (beta) Search your code and get back the owning team, the service, and whether the match is running in production. Behind a feature flag, and it needs a search backend you turn on yourself.

• Integration:Crawler GitLab, cloud and self-managed, plus Gitea.

• Integration:Cloud Microsoft Azure.

• Governance Risk classification, incidents with RACI, and scheduled reminders. Risk includes the share of each service’s changes that shipped without a person reviewing them.

• Crawler Performance work on repository discovery.