Skip to main content

MCP Analyst

Learning goals​

Use an external tool as structured context for an AI response.

Quick reference: Level 4; advanced; output type tool_augmented_brief; estimated cost: tool and LLM provider dependent.

Architecture​

Text alternative: The graph contains FlowBeads/input, LLMBeads/MCPTool, LLMBeads/LLMProcess, FlowBeads/output. It accepts public input and returns a public result.

Prerequisites​

None.

Connections​

  • One configured MCP server and tool

Secrets​

  • MCP credentials when required
  • One supported LLM secret

Build and run​

  1. Select an MCP server and tool.
  2. Select an LLM secret.
  3. Run and inspect tool and LLM events separately.

Use this fixture in the Test panel:

Tool task
Inspect the configured source and summarize the most important findings.

Expected result​

  • MCP invocation is audited
  • normalized tool output reaches LLM context

Open the persisted run and verify the node timeline. Browser sinks appear in the Inspector; public output appears in the run result when enabled.

Side effects and cost​

This Braid has no external side effect.

Cost: tool and LLM provider dependent. DataBraid stores summaries and structured errors, not secret values.

Extensions​

  • Change the fixture and compare the node-level timeline.
  • Add a Data Probe before the final output.
  • Save a known-good fixture for regression testing.