Usage — API / CLI / MCP

leanix-agent exposes the same capability three ways: as MCP tools an agent calls, as a Python API (LeanixApi and GraphQL) you import, and as a CLI. The standardized agent-package pattern and the concept registry are in Architecture.

As an MCP server

Once deployed, the server registers action-routed tools across 30+ LeanIX service domains. Each domain is toggled by its own LEANIX_*TOOL environment flag, and the dynamic toolset filter keeps the active surface lean.

Group	Domains
Core	graphql, universal_api, instance_graph, leanix_pathfinder, leanix_metrics, leanix_mtm
Discovery	leanix_discovery_saas, leanix_discovery_sap, leanix_discovery_ai_agents, leanix_discovery_linking_v1, leanix_discovery_linking_v2
Integrations	leanix_integration_api, leanix_integration_collibra, leanix_integration_servicenow, leanix_integration_signavio, leanix_apptio_connector
Catalog & data	leanix_reference_data, leanix_reference_data_catalog, leanix_inventory_data_quality, leanix_technology_discovery
Operations	leanix_automations, leanix_documents, leanix_impacts, leanix_navigation, leanix_poll, leanix_survey, leanix_synclog, leanix_todo, leanix_transformations, leanix_webhooks, leanix_storage, leanix_managed_code_execution, leanix_ai_inventory_builder

Example agent prompts that map onto these tools:

• "Search the inventory for applications named like 'CRM'" → Pathfinder FactSheet search
• "Introspect the workspace meta-model before I mutate a FactSheet" → leanix_discover_meta_model
• "List the KPIs tracked for application <id>" → Metrics
• "Compile the current model and return its digest and counts" → leanix_generate_instance_ontology
• "Run a governed delta into the operational graph" → leanix_sync_instance_to_graph

As a Python API

LeanixApi is a REST facade over the LeanIX Pathfinder API. Build a client straight from the environment with get_client(), or construct one directly:

from leanix_agent.auth import get_client

api = get_client()        # reads the runtime-projected AgentConfig values

# Reads
factsheets = api.get_factsheets()              # one bounded FactSheet page
factsheet = api.get_factsheet(id="<guid>")     # a single FactSheet by id

Construct the client explicitly only inside a trusted child after its supervisor has materialized the selected runtime references:

from leanix_agent.api.api_client_leanix import LeanixApi

api = LeanixApi(
    base_url=runtime_workspace_url,
    token=runtime_access_token,
    is_oauth=True,
)
factsheets = api.get_factsheets()

GraphQL

The Pathfinder GraphQL endpoint is exposed through the GraphQL client for flexible queries and meta-model introspection:

from leanix_agent.auth import get_graphql_client

gql = get_graphql_client()

result = gql.query(
    """
    query {
      allFactSheets(first: 5) {
        edges { node { id name type } }
      }
    }
    """
)

Both clients resolve the selected AgentConfig TLS profile. Certificate and hostname verification cannot be disabled; private CA, mTLS, and proxy settings belong in the external profile rather than source or tool arguments.

Live ontology and governed graph synchronization

leanix_generate_instance_ontology compiles the current privacy-safe live model to deterministic OWL, SHACL, and SKOS. leanix_sync_instance_to_graph drains bounded cursor pages, tolerates usable GraphQL partial responses, retries a page with a minimal safe selection when an optional field fails, replaces external identities with stable opaque references, and writes only through native atomic ChangeEnvelope operations. The sync requires an ambient verified GraphSession with kg:write; it never constructs graph authority from tool arguments.

Every REST or GraphQL mutation requires allow_mutation=true on that individual request. There is no process-wide mutation bypass.

As a CLI

The MCP server itself is the primary CLI (leanix-mcp); the optional agent server ships as leanix-agent:

# MCP server
leanix-mcp --transport streamable-http --host 127.0.0.1 --port 8000

# A2A agent server
leanix-agent --provider <configured-provider> --model-id <configured-model>

See Deployment for the full transport matrix, the agent server, and client registration.