Skip to content

MCP and agent access

Your agent can read the definitions, not your credentials

Studio runs a Model Context Protocol server, so an assistant can explore the ontologies you have published: list them, search classes and properties, walk the hierarchy, and read the RDF source of any one of them. It is read-only by construction. It cannot run a Transformer, and it cannot write to a graph.

Book a demo Why context, not chunks
Serving path 16:9
Sources Build Store Consume
Kafka & Files
SQL
RESTful endpoints
graph.build Transformers
Ontology
Writers
Your graph database SPARQL · Gremlin · Cypher
Agent any model, any framework
MCP

MCP on the Studio Node: an assistant reads the published ontology, read-only, with no source credentials handed to the model

The boundary

Definitions are a much smaller thing to hand over

Most of the risk in an agent programme is not the model. It is what the model is allowed to reach. An assistant that can read what a Supplier means does not need a connection to the system the suppliers came from.

Definitions, not records

What the surface serves is the ontology: classes, properties, hierarchies and the RDF behind them. Your instance data stays in your graph database, behind whatever access control you already put there.

Read-only by construction

There is no write path through it. It cannot create a model, it cannot run a Transformer, and it cannot load anything into a database.

Only what you published

The list an assistant can see is the ontologies published on your instance. Work that has not been published is work it cannot ask about.

On your Studio Node

The endpoint is part of your own deployment, on your network, alongside the rest of the platform, and it takes a bearer token your instance issues. Nothing is routed through us.

agent config json
{
"mcpServers": {
"graphbuild": {
"url": "http://localhost:9601/mcp",
"headers": {
"Authorization": "Bearer gb_"
}
}
}
}
// point it at the Studio Node, not the Studio UI
// the token is issued by your instance and scoped to it
// no source system credential is involved

What it exposes

The ontology, and a way to walk it

The assistant is not handed a query language and told to guess a schema. It is given the definitions, and tools for reading them.

the surface

Ask which property a column should map to

Alongside the tools there are ready-made workflows. `map-source-to-ontology` takes a field name such as customer_name, pulls keywords out of it, searches the ontology and recommends the class or property it should map to. `explore-ontology` walks a whole ontology from the top down. `explain-class` assembles the parents, children, hierarchy and full RDF description of one class and explains how to use it.

The consequence: the modelling question a new starter would ask a colleague is one an assistant can now answer from your own definitions, rather than from what it read on the internet.

mcp · tools

Supplier 4,214 nodes

describe_ontology label, comment, where to start

list_root_classes top-level classes

search_ontology keyword, classes or properties

get_hierarchy ancestors or descendants

get_resource_cbd full RDF, as n-triples

validates · 0 constraint violations

Point your assistant at your own definitions

Bring the client you already use. We will stand up a Studio Node over a model of your domain and let it read.

Book a demo Talk to us
  • Runs in your account
  • Read-only, and no source credentials leave your side
  • Works with Cursor, Claude and any MCP-capable client