run_flow

Execute a Langflow flow with inputs and retrieve execution results.

Purpose: Run a workflow and receive its outputs. Essential for testing flows, integrating workflows into applications, and automating tasks. Supports both synchronous execution and streaming for real-time responses from LLM-based flows.

Parameters:

• flow_id_or_name (required, string): Flow identifier (UUID) or flow name (case-sensitive)

• input_request (required, object): Execution configuration containing:

  • input_value (optional, string): Primary input data for the flow (e.g., user message, data to process)

  • output_component (optional, string): Specific output component to return

  • output_type (optional, string): Expected output format (e.g., "chat", "text", "json")

  • input_type (optional, string): Type of input being provided (e.g., "chat", "text", "json")

  • session_id (optional, string): Session ID for conversation continuity

  • tweaks (optional, object): Component-specific parameter overrides (key: component_id, value: parameter object)

• context (optional, object): Request context passed alongside the input request

• stream (optional, boolean, default: false): Streaming is currently rejected by this MCP client because streamed responses are not JSON serializable

Returns: RunResponse object containing:

• outputs (array): Flow execution results from output components

• session_id (string): Unique session identifier for conversation continuity

• Additional fields vary by flow configuration (message, data, artifacts)

Usage Examples:

1. Simple execution: { flow_id_or_name: "my-chatbot", input_request: { input_value: "Hello" } }

2. With output type: { flow_id_or_name: "data-processor", input_request: { input_value: "data", output_type: "json" } }

3. With context: { flow_id_or_name: "llm-chat", input_request: { input_value: "Explain AI" }, context: { tenant: "acme" } }

4. With tweaks: { flow_id_or_name: "flow-uuid", input_request: { input_value: "test", tweaks: { "component-id": { temperature: 0.7 } } } }

Best Practices:

• Use flow ID (UUID) for production; names may change or conflict

• Build flow first with build_flow to ensure it's valid before execution

• Leave stream false; this MCP client currently returns JSON tool responses only

• Cache session_id for multi-turn conversations to maintain context

• Set appropriate timeout values for long-running flows

• Test flows with various inputs before production deployment

• Use tweaks sparingly; prefer configuring flows in Langflow UI

Common Errors:

• "Flow not found": Verify flow_id_or_name exists using list_flows

• "Flow has not been built": Run build_flow before executing

• "Invalid input configuration": Ensure input_request matches flow's expected inputs

• "Component configuration error": Check tweaks object matches component IDs and parameter names

• "Execution timeout": Flow took too long; increase timeout or optimize flow

• "Missing required input": Flow expects input_value but none provided

Related Tools:

• build_flow: Compile flow before execution to validate structure

• get_flow: Inspect flow structure to understand required inputs

• get_build_status: Check build completion before running flow

• trigger_webhook: Alternative execution method via webhook endpoint

• list_flows: Find available flows to execute