logo
AgentLed

KG API

Query entities, traverse relationships, and ingest data into the Knowledge Graph through MCP tools and REST endpoints.

MCP Tools

The primary way to interact with the Knowledge Graph is through MCP tools, available from any connected client.

query_kg_edges

Query relationship edges in the graph. Filter by source entity, target entity, edge type, or metadata.

// Query all SCORED edges for a specific investor
{
  "source_name": "Sequoia Capital",
  "edge_type": "SCORED",
  "limit": 50
}

// Query all edges between two entities
{
  "source_name": "Acme Corp",
  "target_name": "Series A Fund",
  "limit": 10
}

get_scoring_history

Retrieve the full scoring history for an entity across all workflow executions.

// Get scoring history for an investor
{
  "entity_name": "Sequoia Capital",
  "edge_types": ["SCORED", "PROCEED_TO_IC", "HOLD"]
}

Returns timestamped scores with rationale, enabling trend analysis and calibration.

get_knowledge_rows

Fetch rows from a Knowledge List. Lists are structured datasets (CSV-like) stored in the workspace.

// Get all rows from the "investors" knowledge list
{
  "list_key": "investors",
  "limit": 100,
  "offset": 0
}

get_knowledge_text

Retrieve text content from a knowledge entry. Useful for documents, notes, and unstructured data.

// Get text content for an ICP definition
{
  "key": "icp-definition"
}

Ingestion Endpoints

Push data into the Knowledge Graph from external systems. All endpoints accept JSON payloads.

EndpointPurpose
/ingest/historicalBatch import historical entities (companies, contacts, deals)
/ingest/knowledge-dataSync workspace Knowledge List rows to graph entities
/ingest/insightStore AI-generated insights linked to entities
/ingest/execution-knowledge-linkLink execution results to knowledge nodes with scored edges

Within AI Steps

AI steps in workflows automatically receive four KG tools. No configuration is needed — the tools are injected based on the workspace's Knowledge Graph connection.

// Inside an AI step prompt, the model can call:
kg_search("fintech investors who led Series A in 2025")
kg_traverse("Sequoia Capital", "INVESTED_IN")
kg_nodes("investor")
kg_write({
  insight: "Sequoia shifted focus to climate tech in Q3",
  entity: "Sequoia Capital",
  category: "market_signal"
})

When the Knowledge Graph is unavailable, tools return empty results instead of errors, so workflows degrade gracefully.

Next Steps