> ## Documentation Index
> Fetch the complete documentation index at: https://alphacastio.mintlify.site/llms.txt
> Use this file to discover all available pages before exploring further.

# MCP Examples

> End-to-end prompt walkthroughs that combine multiple Alphacast MCP tools to solve realistic discovery and data-fetching tasks.

These walkthroughs show the kinds of prompts that work well against the Alphacast MCP server. Each one uses several tools in sequence — your client (Claude Code, Codex, etc.) decides the order based on the prompt, but the underlying tool calls are noted so you can debug if something looks off.

<Note>
  New to phrasing requests for the MCP? Start with [Writing better prompts](/mcp/prompting) — the five guidelines, copy-paste templates, and the preview-before-commit pattern for write actions.
</Note>

## Find inflation series and download them

> "Find the latest Argentina inflation series in the Alphacast catalog, then download the most recent one as CSV."

What the model does:

1. **`search_catalog`** with `query: "Argentina inflation"` and `asset: "datasets"`.
2. Picks the top result and confirms with **`get_dataset`** to verify columns and date range.
3. Calls **`download_dataset`** with `format: "csv"` to get the rows.

Tips:

* Add `excludeDeprecated: true` to the search if you want only currently-maintained series.
* If your model is going to summarize the data instead of save it, ask for `format: "json"` so it can iterate over rows directly.

***

## Profile a dataset before downloading

> "Profile dataset 12345 and tell me if it's worth downloading — what entities does it cover and how fresh is it?"

What the model does:

1. **`get_dataset_profile`** with `datasetId: 12345` and `sampleRows: 5`.
2. Reports the entity breakdown, date range and frequency, total row count, and a small sample — without downloading the full dataset.

Tips:

* Profiling is much faster than downloading when you just need to decide whether a dataset is relevant.
* Combine with `search_catalog` to profile multiple search results in sequence before committing to a full download.

***

## Author a chart from an existing pipeline

> "Show me how to configure a line chart for pipeline 4821, validate the config, and add it as a chart-data step."

What the model does:

1. **`get_pipeline`** with `pipelineId: 4821` to see the step list and output columns.
2. **`get_grapher_config_reference`** with `chartType: "LineChart"` to get the minimal config template and text length limits.
3. Fills in the template using column names from the pipeline's output.
4. **`validate_grapher_config`** with `pipelineId: 4821` and `stepOrder` to confirm the config is valid against the actual upstream data.
5. **`add_pipeline_step`** to append a `chart-data` step with the validated config.

Tips:

* Always validate before adding the step — the validator catches missing column references and invalid option values that would fail silently otherwise.
* Use **`get_chart_details`** on an existing chart to clone its style rather than building from scratch.

***

## Catalog discovery for a research question

> "I want to research how Brazil's central bank rate compares to inflation. Find the relevant series across providers and tell me which combinations are available."

What the model does:

1. **`list_providers`** with `query: "Brazil"` to surface BCB, IBGE, etc.
2. For each candidate provider: **`browse_provider`** to find the relevant categories.
3. **`search_provider`** with focused keywords (`"selic"`, `"IPCA"`).
4. **`get_series`** on the top candidates to confirm date range and frequency.
5. Reports the combinations that align in frequency and date coverage.

Tips:

* Use `search_provider` to narrow results before fetching full series data — it keeps the conversation fast and avoids token bloat.
* If two providers carry the same series, prefer the one whose `license` field is most permissive for your use case.

***

## Compare a series across multiple providers

> "Find unemployment rate series for the United States from FRED, BLS, and OECD. Pull a recent preview from each and tell me how they differ."

What the model does:

1. **`search_provider`** on `fred` with `query: "unemployment rate"`, then on `bls` and `oecd` with similar queries.
2. **`get_series`** on the top candidate from each provider.
3. Compares frequency, seasonal adjustment, and date coverage across the three.

Tips:

* The same indicator can have different methodology across providers (e.g., seasonally adjusted vs. raw, monthly vs. quarterly). The metadata returned by `get_series` calls this out — read it before treating the series as interchangeable.