FlowiseAI/Flowise · error · Error

${getErrorMessage(e)}

Error message

${getErrorMessage(e)}

What it means

Inside Agent.handleToolCalls, each tool invocation is wrapped in try/catch. On failure the error is logged, recorded into usedTools with the error message, streamed via sseStreamer, and re-thrown verbatim through getErrorMessage(e). This propagates the original tool failure (not wrapped) up to Agent.run's catch, which wraps it as 'Error in Agent node'.

Source

Thrown at packages/components/nodes/agentflow/Agent/Agent.ts:2361

                    const errMsg = getErrorMessage(e)
                    let toolInput = toolCall.args
                    if (typeof errMsg === 'string' && errMsg.includes(TOOL_ARGS_PREFIX)) {
                        const [_, args] = errMsg.split(TOOL_ARGS_PREFIX)
                        try {
                            toolInput = JSON.parse(args)
                        } catch (e) {
                            console.error('Error parsing tool input from tool:', e)
                        }
                    }

                    usedTools.push({
                        tool: selectedTool.name,
                        toolInput,
                        toolOutput: '',
                        error: getErrorMessage(e)
                    })
                    sseStreamer?.streamUsedToolsEvent(chatId, flatten(usedTools))
                    throw new Error(getErrorMessage(e))
                }
            }
        }

        // Return direct tool output if there's exactly one tool with returnDirect
        if (response.tool_calls.length === 1) {
            const selectedTool = toolsInstance.find((tool) => tool.name === response.tool_calls?.[0]?.name)
            if (selectedTool && selectedTool.returnDirect) {
                const lastToolOutput = usedTools[0]?.toolOutput || ''
                const lastToolOutputString = typeof lastToolOutput === 'string' ? lastToolOutput : JSON.stringify(lastToolOutput, null, 2)

                if (sseStreamer && !isStructuredOutput) {
                    sseStreamer.streamTokenEvent(chatId, lastToolOutputString)
                }

                return {
                    response: new AIMessageChunk(lastToolOutputString),
                    usedTools,

View on GitHub (pinned to abe4a8601a)

Solutions

1. Identify which tool failed from the streamed usedTools event (selectedTool.name) and its error field.
2. Open that tool node and test it in isolation; fix its config/credential/endpoint.
3. If the tool is external, verify connectivity and auth from the Flowise host.
4. Re-run the agent once the failing tool is corrected; consider marking the tool optional if appropriate.

async function safeToolInvoke(tool, input) {
  try { return { ok: true, value: await tool.invoke(input) } }
  catch (e) { return { ok: false, error: e } }
}
// pre-flight: validate tool is init'd and credentialed
for (const t of toolsInstance) {
  if (typeof t.invoke !== 'function') throw new Error(`Tool ${t.name} is not invokable`)
}

function isInvokableTool(t) { return !!t && typeof t.invoke === 'function' }

try { /* agent.run */ }
catch (e) {
  const m = e instanceof Error ? e.message : String(e)
  // the streamed usedTools event carries the failing tool name + error;
  // match on tool name to route to the right remediation
  if (/401|Unauthorized/.test(m)) { /* refresh tool credential */ }
  throw e
}

When it happens

Trigger: A configured tool node throwing during .invoke/.call: HTTP tool hitting a bad endpoint, calculator tool bad input, RAG/tool credential missing, custom tool code throwing, MCP tool transport error, tool input args failing JSON parse inside the tool.

Common situations: Tool sub-node missing its credential; tool API endpoint down or returning non-200; tool input schema mismatch; version skew between tool schema and tool implementation; network/firewall blocking the tool's outbound call.

Related errors

• Agent needs to have a function calling capable models.
• Error executing tool. Tool: ${tool.name}. Thread ID: ${threa
• Tool ${call.name} not found.
• Model is required
• Invalid human input type. Expected 'proceed' or 'reject', bu

AI-assisted analysis of FlowiseAI/Flowise@abe4a8601a (2026-08-12). Data as JSON: /api/errors/430ca3a92957dabf. Report an issue: GitHub.