LSCP sits between autonomous models and your systems, enforcing rigorous boundaries on what agents can extract, observe, or execute against a Nexum-compiled graph.
Requests outside the bound schema are dropped before they ever reach your systems.
LSCP defines strict semantic limits for what a model is permitted to generate. When a model attempts to synthesize a node, field, or relationship that falls outside the explicit, pre-approved target schema, the payload is rejected before it can propagate—no silent data creation, no schema drift.
This is what stops "runaway generation": an agent can only ever produce what it was explicitly bound to produce.
{
"request": "extract:invoice.total",
"schema_bound": true,
"attempted_field": "invoice.internal_notes",
"decision": "REJECTED",
"reason": "outside_bound_schema"
}
{
"mcpServers": {
"lscp-builder": { "command": "npx", "args": ["@x12i/lscp-builder"] },
"lscp-runtime": { "command": "npx", "args": ["@x12i/lscp-runtime"] }
}
}
LSCP exposes built-in Builder and Runtime Model Context Protocol (MCP) servers. Builder handles schema definition and bound-extraction configuration; Runtime handles live enforcement against agent requests.
Both integrate instantly with major orchestration frameworks—no custom connectors, no bespoke auth flows.
Every node synthesized by an agent carries an immutable hash pointing to its exact source, timestamp, and extraction methodology. Nothing enters the graph — or leaves it in a response — without a verifiable chain of custody.
That gives compliance and security teams total visibility into what an agent saw, extracted, and acted on.
{
"node": "invoice:4471.total",
"value": "$12,400.00",
"source_hash": "sha256:7c1e...",
"extracted_at": "2026-08-14T09:12:03Z",
"extraction_method": "bound_field_read",
"lineage": ["nexum:draft#88a2", "openapi:billing.yaml"]
}
LSCP governs access to a graph that Nexum compiles from your documentation, specs, repos, and records.
See Nexum's Graph Architecture →