MCP Integration

Connect AI agents to UAPF packages via the Model Context Protocol — structured, authenticated access to processes, decisions and runtime execution.

Engine & MCP Status
UAPF Engine
Not configured  offline
MCP WebSocket
Not configured — set UAPF_MCP_WS_URL env var
Config method
UAPF_ENGINE_URL env var  ·  or uapf/config/uapf.ini
MCP Tools

The UAPF Engine exposes these tools over MCP. AI agents can discover them via the describe tool or by reading the manifest.

describe

Returns the capabilities of the connected UAPF Engine — version, supported standards, available packages.

list

Lists all packages available in the registry, with ID, name, version and level metadata.

filter?level?
run_process

Initiates a BPMN process instance with the provided variables. Returns instance ID and initial state.

packageIdprocessIdvariables
evaluate_decision

Evaluates a DMN decision or decision table against the supplied input context. Returns the output values.

packageIddecisionIdcontext
resolve_resources

Resolves resource bindings for a process or task — role assignments, system integrations, service endpoints.

packageIdprocessId?taskId?
get_artifact

Returns the raw BPMN, DMN or CMMN model XML, or the manifest JSON, for a given package and artifact identifier.

packageIdkindartifactId?
validate

Runs deep validation against JSON Schema and structural rules for a package or uploaded manifest. Returns issues list.

packageId?
Agent Connection Flow
1

Register the agent

POST to the UAPF Engine agent registry with the agent's DID and name. Receive a one-time session key.

2

Open a WebSocket session

Connect to the MCP WebSocket endpoint. Present the session key in the handshake headers.

wss://your-engine-host/mcp/ws Authorization: Bearer <session-key>
3

Discover available tools

Call describe to get the engine manifest, then list to enumerate packages.

{ "tool": "describe", "input": {} } → { "engine": "uapf-engine", "version": "1.x", "packages": 4 }
4

Invoke processes & decisions

Use run_process and evaluate_decision to execute governed logic inside the agent's reasoning loop.

{ "tool": "evaluate_decision", "input": { "packageId": "retail-credit-approval", "decisionId": "eligibility-rules", "context": { "creditScore": 680, "income": 52000 } } }
🤖
Using with Claude & MCP

The UAPF Engine MCP server can be registered in Claude Desktop or any MCP-compatible host. Once connected, Claude can discover available packages, reason about process logic and invoke governed decisions — replacing ad-hoc instructions with structured, auditable process execution.

Engine reference implementation ↗