Know what your agents do.
Give the work a control plane.
Centurian helps organizations see what their AI agents are doing, set shared rules, track costs and review the results.
An agent fleet can grow before anyone can see it.
One team uses an agent in a cloud platform. Another builds its own. A vendor introduces a third. Each has an owner, tools, costs and consequences, but the organization still needs a shared view of the work. Centurian starts with that operational problem.
Keep local ownership. Add shared visibility.
I designed Centurian as a shared management layer for AI agents. It uses Model Context Protocol (MCP) for connections and a federated design, meaning each team keeps local control. Teams keep their agents, internal logic and local policies. A shared layer supplies fleet visibility, organization-level rules, evaluation, cost aggregation and audit evidence. Transportation and logistics is the intended starting market.
Five jobs, one shared record
Connect covers integrations and agent registration. Govern covers ownership and policy. Measure examines agent trajectories and evaluation results. Cost brings per-agent spending and budgets into view. Prove organizes audit evidence and compliance-framework reporting. The product logic is to connect these jobs around the same agent and action history.
The policy decision is federated.
An action passes through two sets of rules: the agent owner’s local rules and the organization’s global rules. Both must allow an action. Global policy uses normalized attributes such as the action, data classification, destination and reversibility, so the rule can apply across different agent platforms.
The implementation and its boundaries
The application uses TypeScript, Next.js, tRPC, Prisma, PostgreSQL, Redis and Clerk. Open Policy Agent (OPA) supplies the policy layer. A shared core supports an MCP server and a planned REST gateway, with common contracts and scoped access.
MCP is the working integration surface. An equivalent REST gateway remains planned, so REST support should not be assumed. Audit records help people examine what happened; they do not establish regulatory certification.
Start with one agent. Keep the enterprise path coherent.
The product design gives an individual, a team and an organization different views of the same system. A developer should get value from understanding one agent before an enterprise asks to govern a fleet. My work combines that adoption path with the architecture needed for shared policy and evidence.
A few useful answers
What is Centurian?
Centurian is Adnan Khan’s independent MCP-first control-plane project for AI agents. Its product model combines Connect, Govern, Measure, Cost and Prove around shared agent and action records.
What technology does Centurian use?
The repository documents TypeScript, Next.js, tRPC, Prisma, PostgreSQL, Redis, Clerk and Open Policy Agent. MCP is the primary integration surface; the transport documentation marks REST parity as planned.
Is Centurian certified for regulated industries?
This case study does not claim regulatory certification. The product design supports policy and audit evidence; certification and verified customer outcomes are separate matters.
Sources and scope
Based on Adnan’s Centurian product index, federated architecture and transport-adapter documentation, and package manifest reviewed in October 2026. The documents distinguish MCP capabilities from planned REST parity. This case study does not independently verify production deployment, certification or customer results.
