Jev MCP — the next level of MCP
v5 → v7 The next level of MCP servers

Stop listing tools. Start guiding agents.

HAPI MCP turns the OpenAPI and Arazzo specs you already have into endpoint and workflow tools, then plans the path so any assistant knows which one to call next. No business logic re-coded.

Jev MCP builds on HAPI MCP by adding deterministic planning capabilities, ensuring that your AI agents follow predictable workflows even in complex scenarios.

$ npm install -g @rebelion/hapi-mcp
Endpoints and workflows, served as tools
openapi.yaml · 48 endpoints refund.arazzo.yaml · 3 workflows
HAPI MCP generated from your specs
endpointlistOrders
endpointcreateRefund
workflowrefundLatestOrder(email)
One deterministic call, for every agent
Code mode needs an agent that codes. Workflows don't.
AgentCode modeHAPI Workflow
{{ ag.name }} {{ ag.cm }} ✓ one call
Defined once in Arazzo
refundLatestOrder(email)
Knows which tool to call next
findCustomer → listOrders → ask the user → createRefund
AssistantWhich order should I refund — #1042 or #1043?
#1043, please.
Paused Refund $42.00 on #1043? Approve
Clawne mascot
{{ e.v }}
{{ e.label }}
Where MCP has been Where HAPI takes it
What's new

Three levels beyond the tool list

One tool per endpoint, discovery layers and code mode each fixed part of the problem. HAPI MCP picks up where they stop.

v5HAPI MCP

Endpoints and workflows, served as tools

Point HAPI at your OpenAPI and Arazzo files. Raw endpoints stay available for flexibility; workflows bundle every step into one ergonomic operation.

No business logic re-coded Brownfield or greenfield
openapi.yaml *.arazzo.yaml
HAPI
endpoint tools workflow tools
Your existing API — unchanged
{{ ag.name }} {{ ag.cm }} ✓ workflow
v6Beyond coding agents

Code mode's result, without the code

Code mode is powerful, but only for agents that can write and run scripts in a sandbox. HAPI Workflows give assistants and AI teammates the same outcome in one deterministic tool call.

Same steps, every run Defined in Arazzo
v7HAPI Capability Planning

Dynamic paths, with guardrails built in

HAPI uses Jev's System One to classify every tool, so assistants understand the path through a workflow, ask for missing details, and pause before important changes.

LookupSafe reads the agent can chain freely.
ActionWrites that need the right inputs first.
Needs approvalImportant changes wait for a human.
AssistantFound Ana's account and 2 recent orders.
AssistantWhich order should I refund — #1042 or #1043?
#1043, please.
Paused Refund $42.00 on #1043? ✓ Approved
An API with 50 MCP tools is useful.
An AI that knows which one to call next is transformative.
Less guessing Better conversations More trustworthy automation
Install

Run it your way

A single CLI, the same specs, wherever your agents live.

{{ termTitle }}
{{ l.prompt }}{{ l.text }}
# then serve your spec
$ hapi init my-mcp-server
$ hapi serve ./openapi.yaml

Take your MCP server to the next level

Keep your API. Add workflows and planning. Give every agent a clear, safe path.

Clawne mascot