FERC Open Data MCP Server

FERC Open Data

Public datasets published on data.ferc.gov. MCPBundles supplies the FERC API key server-side, so users can enable the server without entering a credential.

Workflow

1. Catalog overview when the user asks what is available overall.
2. Find datasets by keyword when the user names a topic (MBR seller authorization, PPA, generator asset, annual charges, company registration, NEPA).
3. Scan dataset rows when the user names a company, docket, or counterparty and you already know the dataset_id. The tool examines only a bounded slice of the table and returns slice.fraction_examined — quote that coverage; never treat partial matches as a full cohort.
4. List datasets when you need the full flat index without a search term.
5. List data assets when you need the nested asset tree with program-office context.
6. Get dataset details for row_count, freshness, and whether a dictionary exists.
7. Get dataset dictionary before wide pulls when available.
8. Get dataset data with limit and offset for deliberate pagination or sampling.

Dataset IDs can change over time. Always resolve the current ID from find/list before details, dictionary, or data.

Datacenter / power-market research

For seller authorizations, PPAs, and generator assets, search Market-Based Rate Database datasets (MBR Authorizations, Entities to PPAs, Gen Assets). Company identifiers live under Company Registration. There is no way to rank sellers nationally by megawatts or contract value in this API — only paginated table reads.

What is not in this API

Electric Quarterly Reports (EQR) and FERC Form 1 utility financials are not exposed through data.ferc.gov yet. If the user asks for EQR or Form 1, say so honestly — do not substitute unrelated datasets.

Gotchas

• No server-side row filters on the data endpoint — only pagination via limit and offset.
• Row scan is slice-bounded — ferc_search_dataset_rows examines only part of large tables and returns coverage_note with rows_examined vs row_count. Do not aggregate or rank from partial scans.
• Large tables (>5000 rows) require a field from ferc_get_dataset_dictionary before row scan.
• Rate limit: 1,000 requests per hour (rolling). Prefer find + details + small page reads.
• Dictionary gaps: not every dataset publishes a dictionary; check contains-dictionary in details first.
• Large tables: MBR and registration datasets can have thousands of rows — paginate deliberately.