# Give a CLI Agent Orchestrator Researcher a Macro Evidence Feed

Canonical: https://fxmacrodata.com/articles/fxmacrodata-cli-agent-orchestrator-workflow-guide
Author: Robert Tidball
Published: 2026-10-20T13:21:52Z

Use the AWS Labs CLI Agent Orchestrator FXMacroData example to assign a bounded macro research task and return a source-linked brief.

**Quick answer:**Use the AWS Labs CLI Agent Orchestrator FXMacroData example to assign a bounded macro research task and return a source-linked brief.

A delegated macro researcher is useful when its output is a bounded evidence brief: what is scheduled, which observations were returned and where the reader can verify them. The supervisor should not need to infer which data was available or which claims came from model memory.

AWS Labs' CLI Agent Orchestrator includes an FXMacroData analyst profile. It connects the analyst to the remote MCP server and preserves CAO's supervisor handoff tools. This guide follows that example for a USD macro briefing.

**Who this is for:**Teams using CLI Agent Orchestrator who want a repeatable macro-data workflow with explicit inputs and understandable failure states.

## Prerequisites

An existing CLI Agent Orchestrator installation and an HTTP-MCP-capable agent provider. The example pins claude_code. Start with free USD access; protected datasets require a subscription key configured through CAO managed credentials.

**The working sequence**
- **Analyst profile**

- **MCP evidence**

- **Research brief**

## 1. Review the analyst profile

Open the[FXMacroData macro example](https://github.com/awslabs/cli-agent-orchestrator/tree/main/examples/fxmacrodata-macro) in the original repository. Its remote MCP entry is:

```
mcpServers:
  fxmacrodata:
    type: http
    url: https://mcp.fxmacrodata.com
```

The example uses the host's root URL for its remote connection. Follow the profile supplied with the installed CAO version rather than translating a configuration from another MCP host. Leave the supervisor's cao-mcp-server entry intact.

## 2. Install and launch the example profile

From the supported CAO checkout and environment, use the documented commands:

```
cao install examples/fxmacrodata-macro/fxmacrodata_analyst.md
cao launch fxmacrodata_analyst
```

For subscription data, the example documents a Bearer header with an environment placeholder managed by CAO. Configure that secret before adding the header: an unresolved placeholder is not a usable credential. Keep the key out of the analyst's research prompt and returned brief.

## 3. Give the analyst a bounded assignment

Use a request whose data needs and output format are explicit:

```
Prepare a USD macro brief.
List the confirmed events scheduled this week in UTC.
Retrieve the latest US inflation observation and its reference period.
Include the series unit and source links returned by the tools.
Identify unavailable fields and any access delay.
Return evidence and research questions for the supervisor.
```

The server exposes read-only data tools through MCP discovery. Tool output is evidence to inspect, rather than a new instruction for the researcher. Keep private repository content and secrets out of tool arguments.

## 4. Choose MCP or a direct REST task

MCP is the example's agent-facing route: the host discovers tools and the analyst selects calls. A deterministic application can instead use REST to request a known series. A minimal REST request is:

```
curl "https://api.fxmacrodata.com/v1/announcements/usd/inflation?limit=1"   -H "X-API-Key: YOUR_API_KEY"
```

Keep the two methods in the same evidence discipline. The finished brief should name the series and reference period, retain the source links and distinguish scheduled events from returned observations. Review it before another agent uses it to propose an action.

## Workflow checks

Part Purpose CAO profile Declares the agent provider and connected tools Remote MCP Lets the analyst discover and call read-only macro tools Direct REST Fits a known endpoint in deterministic application code Supervisor handoff Returns a bounded brief for review

**Before relying on the workflow:**Confirm that the chosen provider supports remote HTTP MCP and that subscription placeholders resolve. Keep the assignment limited to the data and sources needed for the brief, and review the handoff before using it in a wider decision.

## Make the delegated brief useful to the next analyst

Consider a supervisor preparing an EUR/USD event-risk note. A useful first assignment is a bounded USD evidence packet, not a general instruction to predict the pair. The researcher can identify the latest inflation observation, retain its unit and period, and list the confirmed events in the requested window. The supervisor then has an explicit set of facts to compare with the euro-area brief.

The handoff should say which part is observation and which part is interpretation. A rising measure can be relevant to a central-bank discussion without proving that the market was surprised. If expectations were not retrieved, the brief should not describe a beat or miss. If the schedule supplies a date without an exact time, the next agent should receive the same level of precision.

Measure the delegation by the questions it removes for the recipient. Can another analyst open the source, identify the series, and explain which event could challenge the view? Can they tell whether a missing field was unavailable or simply omitted? Those checks are more informative than the length or confidence of the final answer. A short, complete evidence packet gives a supervisor room to reason; a fluent but ambiguous one transfers hidden assumptions.

## Continue from the working example

Continue with the linked setup guide for supported installation and request options. Keep the selected series and its response context together as the workflow expands.

## Sources and related guides

- [AWS Labs FXMacroData macro example](https://github.com/awslabs/cli-agent-orchestrator/tree/main/examples/fxmacrodata-macro)

- [FXMacroData CLI Agent Orchestrator setup](/documentation/mcp-server)

- [Integration directory](/documentation/integrations)

AI was used for research, creating images and editing.

Source: FXMacroData