Last updated
AI Agents & Agentic Systems
Claude Code Agency UK
- Deliverables
- 6
- Process
- 5 steps
- Studio
- Telford, UK
We design, build and support agentic systems on Claude Code and the Claude Agent SDK: agents that work in your codebase, your tools and your processes, with the guardrails a business needs.
In short
Circlemount is a Telford, Shropshire Claude Code agency that builds agentic systems for UK businesses: custom agents on Claude Code and the Claude Agent SDK, MCP connections to your own systems, and safe team rollouts with permissions, hooks and review. Circlemount is independent, not an Anthropic partner, and every engagement starts with a free, tailored proposal.
5.0 Google (17 reviews)£57m client revenue213 clients

Claude Code is powerful. Using it well in a business is a different job.
Claude Code is Anthropic's agentic coding tool. It reads a codebase, edits files, runs commands and connects to other tools, and it works in the terminal, in VS Code and JetBrains, in a desktop app and in the browser. For one developer on one repository it can feel like magic. For a business, the questions change quickly.
Who is allowed to run what? Which systems can an agent reach, and with whose credentials? How do you stop an agent touching production data, leaking a secret or merging code nobody reviewed? How do you turn one person's clever prompt into a repeatable workflow the whole team runs the same way?
That is the work of a Claude Code agency. Circlemount sets Claude Code up as part of how your business builds and runs software: shared project instructions, skills, hooks, subagents and MCP servers, with permissions and review in place from day one, and custom agents built on the Claude Agent SDK where the job goes beyond coding.
Who we help
01
“We want Claude Code rolled out to our dev team safely”
Developers are already experimenting. You want shared settings, sensible permissions, approved MCP servers and visibility before usage spreads further.
02
“We have a process an agent should run”
Triage, reporting, data clean-up, content operations or code maintenance that follows the same steps every time and eats skilled hours.
03
“We need Claude connected to our systems”
Your CRM, database, ticketing or internal APIs, reachable through MCP with the right access, so answers and actions use your real data.
04
“We tried it and it got messy”
Agents that wander, inconsistent results, prompts living in one person's head. You want structure: instructions, skills, tests and review.
05
“We're not technical, but the goal is clear”
You do not want to learn the tooling. You want a working system, explained in plain English, that someone supports.
What We Deliver
01
Claude Code setup and governance
Project and organisation settings, permission rules, managed settings where you need central policy, and a CLAUDE.md that teaches Claude how your codebase and team work.
02
Custom agents and subagents
Specialised subagents for review, testing, research or migrations, and standalone agents on the Claude Agent SDK when the work happens outside a code editor.
03
Skills and hooks
Repeatable workflows packaged as skills your team runs by name, and hooks that format, lint, test or block risky commands around every agent action.
04
MCP integrations
MCP servers that connect Claude to your CRM, database, documents or internal APIs, scoped to the data and actions each agent actually needs.
05
CI and review automation
Claude Code in GitHub Actions or GitLab CI for pull-request review, issue triage and routine maintenance, with people approving what ships.
06
Documentation and handover
Plain-English documentation of every agent, permission and integration, so your team can run, change and audit the system without us.
How It Works
1. Discovery call
We learn what you want automated or accelerated, the tools you use and what must never be touched. No jargon needed.
2. Proposal and scope
A free, tailored proposal: the agents, integrations and guardrails we recommend, what is in and out of scope, and how we will know it works.
3. Build in a safe sandbox
We build against copies, test data or read-only access first, with permissions and hooks in place before an agent touches anything real.
4. Review and roll out
Your team tries it on real tasks. We tune instructions, tools and limits, then roll out to the people and repositories in scope.
5. Support and improve
Models and tools change quickly. We keep agents, settings and integrations current, and extend the system as you find new jobs for it.
Claude Code, Codex or Cursor?
We are tool-agnostic and recommend what fits your team. A factual summary, checked against each vendor's documentation in November 2026; features change often, so we re-check before every recommendation.
| Compared | Claude Code | OpenAI Codex | Cursor |
|---|---|---|---|
| What it is | Anthropic's agentic coding tool | OpenAI's coding agent | An AI code editor with an agent mode |
| Where it runs | Terminal, VS Code, JetBrains, desktop app and browser | CLI, IDE extension, ChatGPT apps and Codex cloud | Its own desktop editor, based on VS Code |
| Project instructions | CLAUDE.md (and can read AGENTS.md) | AGENTS.md | Project rules |
| Connects to tools through MCP | Yes | Yes | Yes |
| Models | Anthropic's Claude models | OpenAI's models | A choice of models from several providers |
| Often chosen for | Agentic work across a repository, terminal and tools; custom agents via the Agent SDK | Teams standardised on OpenAI and ChatGPT | Developers who want AI inside the editor they type in |
What's in your free proposal
Every project starts with a free, tailored proposal. There is no obligation, and it is written for you, not copied from a template.
- Where agents would save your team the most time, and where they should not be used
- The architecture we recommend (Claude Code, the Agent SDK, Codex or a mix) and why
- The MCP integrations and access each agent needs, and nothing more
- Guardrails: permissions, hooks, review steps and how secrets are handled
- What your team needs to learn, and how we hand over
- Scope, timeline and a tailored price
Production software we've built
We are a software and growth studio, not a prompt shop. Circlemount designs, builds and runs production systems, including Circlemount CRM, our own CRM, invoicing, HR, digital signature and proposal platform, and the Portal behind The Study Room, which manages students, tutors, sessions, invoicing and payroll.
Agentic systems get the same engineering discipline as that software: version control, tests, code review, least-privilege access and documentation your team can follow.
Further reading
We Serve Businesses Across the UK
Based in Telford, we provide Claude Code Agency UK to businesses nationally including:
Frequently Asked Questions
Do you build Claude Code agents?
Yes. Circlemount builds Claude Code subagents, skills and hooks for development teams, and standalone agents on the Claude Agent SDK for work outside a code editor, such as triage, reporting or data operations. Every agent gets scoped permissions, tested behaviour and plain-English documentation, so your team knows exactly what it can and cannot do.
What does a Claude Code agency do?
A Claude Code agency designs, builds and supports systems that use Claude Code and Anthropic's Agent SDK inside a business. Circlemount sets up project instructions, permissions, skills, hooks, subagents and MCP integrations, builds custom agents and trains your team, so Claude Code becomes a governed part of how you build and run software.
Are you an Anthropic partner?
No. Circlemount is an independent agency based in Telford, Shropshire, and does not claim any Anthropic or OpenAI partnership or certification. We build on the publicly documented Claude Code, Claude Agent SDK, Codex and MCP tooling, recommend whichever fits your project, and you keep your own accounts, keys and billing with each vendor.
Can you connect Claude to our own systems?
Yes. We build MCP servers, the open standard Claude Code and many other AI tools use to reach external systems, so Claude can read from and act on your CRM, database, documents or internal APIs. Each server exposes only the tools and data an agent needs, with authentication and logging designed in from the start.
Is it safe to let an AI agent work in our codebase?
It can be, with the right setup. We start agents on read-only access or copies, use Claude Code's permission rules and hooks to block risky commands, keep secrets out of prompts and repositories, and require human review before anything merges or reaches production. Safety is designed in at the start, not bolted on afterwards.
Do you work with OpenAI Codex too?
Yes. We are tool-agnostic. Codex, OpenAI's coding agent, runs as a CLI, an IDE extension, in the ChatGPT apps and in Codex cloud, reads AGENTS.md project instructions and can use MCP servers. If your team is standardised on OpenAI, we build on Codex and the OpenAI Agents SDK; many teams use both tools side by side.
How much does a Claude Code project cost?
It depends on scope: a team setup with permissions, skills and hooks is a smaller job than a custom multi-agent system with several MCP integrations. Circlemount does not publish fixed prices for this work. Tell us what you need and you will receive a free, tailored proposal setting out the scope, timeline and price.







