> services
Claude developer for production builds
Senior Claude developers who ship on Claude Code, the Agent SDK, MCP, and Cowork. A fixed-fee discovery sprint, then a production slice in 2 to 4 weeks.
Claude development that ships in weeks, not quarters
Rhode Labs is a senior AI infrastructure studio. We build production software on Anthropic’s Claude: custom agents on the Claude Agent SDK, MCP servers that connect Claude to your tools, Claude Code automation, and Cowork workflows for teams. If you want a Claude developer who has shipped this stack rather than read about it, that is what we do.
We are not a generic AI dev shop that swaps in whatever model is trending. The whole practice is built around Claude and the Model Context Protocol, so the architecture decisions are made by people who know where Claude is strong and where it needs guardrails.
What’s included
- Custom agents on the Claude Agent SDK. The same agent loop that powers Claude Code, programmable in Python or TypeScript, wired to your tools and data.
- MCP servers. Custom servers for your internal APIs, databases, and private systems, plus configuration of existing public servers (Slack, GitHub, Drive) where they already do the job.
- Claude Code automation. Claude running in CI and scripts for code migration, bug triage, doc generation, and other repeatable engineering work.
- Cowork rollouts for teams. Setting up Claude’s desktop agent so non-engineers can run real workflows on the same Agent SDK foundation, with the right MCP connections and permissions.
- Evals and monitoring. Test cases on your real inputs, logging, and permission boundaries, so you can trust the output and catch regressions.
A Claude developer touches four surfaces. Here is how they fit.
Most teams know they want “Claude,” but not which surface solves their problem. The four interfaces share one engine (the Agent SDK) and differ in who drives and where the code runs. This table is how we pick during scoping.
| Surface | Who drives it | Runs where | Best for | Watch out for |
|---|---|---|---|---|
| Claude Code | A developer, interactively | Your terminal | Coding, migrations, repo-wide refactors, ad hoc automation | Great for engineers, not a product surface for end users |
| Claude Agent SDK | Your code | Your servers / CI | Autonomous agents, background jobs, anything programmatic | You own the agent loop, evals, and failure handling |
| MCP servers | The agent, through tools | Anywhere reachable | Giving Claude safe access to tools, data, and private APIs | Auth, transport, and scoping tool permissions tightly |
| Cowork | A non-engineer, interactively | Claude desktop | Team members running real workflows without code | Needs the right MCP connections and permission limits set up first |
The practical rule: Claude Code and Cowork are interactive front doors, the Agent SDK is the same capability as a library you call from code, and MCP is how any of them reach your systems. Anthropic describes the Agent SDK loop as gather context, take action, verify work, with subagents for parallel work and isolated context, plus automatic compaction when a long-running agent approaches its context limit (Anthropic).
How an engagement works
We keep the path short on purpose.
- Discovery sprint (fixed fee, about a week). We map the workflow, pick the surface from the table above, decide build-versus-reuse on MCP servers, and write down the eval cases that define “done.” You get an architecture and a scope, whether or not you continue with us.
- Architecture. Tool boundaries, MCP server design, permission model, and where a human stays in the loop for anything irreversible.
- Deploy a thin slice (2 to 4 weeks). One real workflow in production, wired to your systems, with evals and logging. It is not a demo.
- Iterate and expand. Once a real thing works, we widen scope from there: new tools, more workflows, and tighter automation over time.
That sequence is why we say weeks, not quarters. The slice is small enough to ship fast and real enough to prove the architecture.
Where it pays off
- Internal automation. Multi-step back-office work that spans several tools, run by an Agent SDK agent with scoped MCP access.
- Engineering throughput. Claude Code in CI handling migrations, triage, and the repetitive changes that eat sprint time.
- Connecting Claude to private systems. A custom MCP server is usually the missing piece between a clever prototype and something that touches your real data.
- Team adoption. Cowork lets non-engineers run workflows directly, once a developer has set up the connections and guardrails.
Pricing and engagement
Every engagement starts with a fixed-fee discovery sprint, so you know the architecture and scope before committing to a build. Builds are scoped per project against that plan. We do not quote a flat hourly rate or a one-size monthly retainer, because the right scope depends on which surface you need and how much of your systems the agent has to reach. The discovery output is yours to keep either way.
A note on running costs: starting June 15, 2026, Agent SDK and claude -p usage on Claude subscription plans draws from a separate monthly Agent SDK credit (Anthropic docs). We factor that into architecture so your automation costs stay predictable.
Build with a Claude developer who has shipped the stack
If you have a workflow in mind and want to know what it would take on Claude, start with the discovery sprint. You leave with an architecture and a scope, even if the build happens later.