01
New projects
A paragraph about the problem is enough to start. We reply within two working days, usually with questions rather than a quote.
AI · Service MCP server development
Model Context Protocol servers that give AI agents scoped, audited access to your systems: OAuth scopes per tool, rate limits, and a log of every call.
In short
The Model Context Protocol (MCP) is an open standard for connecting AI assistants and agents to tools and data. An MCP server exposes your system — an API, a database, an internal service — as tools an agent can call. We build MCP servers with scoped OAuth, rate limits and an audit log, so giving an agent access does not mean giving it everything.
- How to engage
- Product build, Security review
- Practice
- AI
Signs this is the right service
- You want assistants such as Claude or your own agents to use an internal system safely.
- Each agent integration is rebuilt by hand with its own auth and error handling.
- You sell a product and customers are asking for an MCP server for it.
- An existing MCP server works, and nobody has reviewed what it is allowed to do.
What you get
Tool design
Tools shaped around tasks rather than endpoints, with clear descriptions, typed inputs and results a model can act on.
OAuth and scopes
Authorization per user and per tool, with read scopes separated from write scopes and tokens that expire.
One server per trust boundary
A server that can read customer records and also post to a chat tool can leak one into the other. We split servers along those lines.
Rate limits and failure handling
Limits per user and per tool, and errors the agent can understand when a downstream system is slow or refuses.
Audit log
Every tool call recorded with who asked, what was sent and what came back, retained for as long as your policy requires.
The process
- 01
Map the surface
Which actions an agent should be able to take, which it should not, and who is allowed to authorize each.
- 02
Design the tools
A short list of well-described tools, tested with a model before any production code is written.
- 03
Build and harden
Authorization, limits, logging and tests, including attempts to make the server do something outside its scope.
- 04
Publish and document
Deployed as a remote server with setup instructions for the clients your users actually have.
MCPOAuth 2.1TypeScriptPythonStreamable HTTPPostgres
MCP server development: common questions
What is the difference between an MCP server and an API?
- An API is written for developers. An MCP server describes tools in a way a model can discover and call, and handles authorization for an agent acting on a user's behalf. It usually sits in front of an existing API.
Should the server run locally or remotely?
- Remote for anything used by more than one person: it can use OAuth, be updated in one place and be audited. Local servers suit developer tools that work on files on one machine.
Is it safe to let an agent call our systems?
- It is as safe as the scopes you grant. We design for the case where the agent is tricked by text it reads, which means it should not hold permissions that would make that dangerous.
Can you review an MCP server we already have?
- Yes. A security review covers the scopes each tool holds, how tokens are handled, what is logged, and whether text returned by one tool can drive a call to another.
06 / Let’s build
Bring us the difficult problem.
02
Careers
Engineers, designers and researchers who like systems that are hard to get right. Send something you built and what you would change about it now.
03
Studio
Based in India and working remotely, with overlap hours agreed up front so your team always has a time to reach us.
04
Read
Guides on AI agents, RAG, MCP, tokenization and chain abstraction, written by the engineers who build them.