Connect an agent in one line
TerraFabric is designed to be operated by software. Agents discover data through a formal ontology, act through typed tools, and spend only within limits a human principal sets.
Add TerraFabric to any MCP client
Point an MCP-capable agent at the site. The same origin serves humans, agents and C2 systems.
- 1
describe_ontology— Learn the vocabulary - 2
search_collections— Find candidate sources - 3
get_collection— Inspect specs, licence, endpoints - 4
search_items— Check scenes over the AOI - 5
get_quote— Price it (24h validity) - 6
place_order— Buy within mandate - 7
get_order— Track to delivery
claude mcp add --transport http terrafabric https://terrafabric.world/mcp
Every capability, machine-readable
Model Context Protocol over streamable HTTP. Tools, resources (ontology, graph, llms.txt) and guided prompts.
The same 15 tools as JSON-Schema function definitions, invokable at POST /v1/agent/tools/{name}.
A compact orientation document: workflow, vocabulary and every collection in one read.
Complete REST description for code-generating agents and API gateways.
Every entity dereferences to JSON-LD with schema.org potentialActions and to Turtle.
Discovery card listing interfaces, skills and spending policy (also at agent-card.json and mcp.json).
Autonomy with a budget
A tf:SpendingMandate is an ODRL policy a principal issues to an agent: a per-order ceiling, a rolling budget and the licence scopes it may buy. Inside the mandate orders confirm instantly; outside it they park in awaiting_approval with an approve action addressed to the human. Tools carry MCP annotations (readOnlyHint, destructiveHint) so clients can gate spend-committing calls.
MCP playground
Call the live MCP endpoint exactly as an agent would. Try the full loop: set_mandate → get_quote → place_order → get_order.
Ids returned by quotes, orders and tasking are chained into later examples automatically.
// Pick a tool and run it against the live MCP endpoint.