Use InfraMap inside your AI tools

Connect

For other clients, add one remote server. Everything else is discovered through the protocol.

https://mcp.inframap.pro/mcp

Clients that take a JSON MCP configuration can use this as is. Steps for each client are under Client setup.

{
  "mcpServers": {
    "inframap": {
      "url": "https://mcp.inframap.pro/mcp"
    }
  }
}

Transport

Streamable HTTP, stateless. There is no separate SSE endpoint.

Sign-in

Your InfraMap account. The client opens the InfraMap sign-in page the first time it connects (OAuth 2.1 with PKCE). No API key.

Plan

Pro or Enterprise. Free accounts can sign in but cannot connect. See plans.

Access

Read-only. All 43 tools are marked read-only and idempotent.

Protocol

Built on the official MCP TypeScript SDK and its supported protocol revisions, including 2025-06-18.

Release manifest

service.json(opens in a new tab)

The server supplies data and calculations. Your AI client writes the answer, so the model you use and its instructions still matter.

Use of the InfraMap MCP server and data is subject to the Terms of Service and Acceptable Use Policy.

Client setup

Remote MCP support depends on the client and, for some, your plan. These steps were checked against each vendor’s documentation on October 25, 2026; menus can change. Claude needs no URL; for every other client, the server URL is the only value you need.

claude mcp add --transport http --scope user inframap https://mcp.inframap.pro/mcp

claude mcp add --transport http --scope project inframap https://mcp.inframap.pro/mcp

codex mcp add inframap --url https://mcp.inframap.pro/mcp

[mcp_servers.inframap]
url = "https://mcp.inframap.pro/mcp"

{
  "mcpServers": {
    "inframap": {
      "url": "https://mcp.inframap.pro/mcp"
    }
  }
}

code --add-mcp '{"name":"inframap","type":"http","url":"https://mcp.inframap.pro/mcp"}'

{
  "servers": {
    "inframap": {
      "type": "http",
      "url": "https://mcp.inframap.pro/mcp"
    }
  }
}

gemini mcp add --transport http --scope user inframap https://mcp.inframap.pro/mcp

gemini mcp add --transport http inframap https://mcp.inframap.pro/mcp

{
  "mcpServers": {
    "inframap": {
      "url": "https://mcp.inframap.pro/mcp"
    }
  }
}

{
  "mcpServers": {
    "inframap": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://mcp.inframap.pro/mcp"]
    }
  }
}

How sign-in works

1. The first time your client calls the server, the server answers that sign-in is required and points to www.inframap.pro.
2. Your client registers itself and opens the InfraMap sign-in page in your browser.
3. Sign in with an email code, then approve the client.
4. The client receives a token that lasts one hour and renews itself for up to 30 days. Each renewal replaces the previous token.

Access follows your plan. If a Pro subscription ends, the client stops connecting at its next renewal. To disconnect, remove the server from your client.

Questions to try

• Find the largest operating wind projects in Texas and show their exact project IDs.
• How many PJM plants have no market price node, and what are the documented reasons?
• Which wind and solar plants saw the largest increase in modeled curtailment from 2024 to 2025?
• Show Barry’s reported generation in 2024, including missing months and source citations.
• What was Moccasin’s day-ahead basis to its hub in June 2025?
• Which datasets can help me research cooling water use, and what do their records mean?

How the tools fit together

Most good answers follow the same order. Capable clients do this on their own after reading the server instructions; it helps to know it when an answer looks thin.

1. Check what exists. get_data_catalog lists every source with its as-of date, coverage and limits. discover_energy_data searches datasets by topic.
2. Find exact projects. Use query_analyst_store for screens across the whole inventory, or search_energy_projects for filtered lists. For a region, call resolve_grid_areas first. A state is never a stand-in for an ISO.
3. Read project evidence. get_project_profile gives a one-call dossier. The generation, price, curtailment, ownership and emissions tools return the detail, with citations.
4. Check definitions. get_energy_reference and get_analysis_guide return reviewed definitions and formulas. explain_analysis_limit says what a missing calculation would need.

Tool reference

43 tools in 10 groups, each with the display title clients show and the name they call. Inputs are listed by name; your client reads each tool’s full schema and description from the server. Most inputs accept null for their default.

Analyst store

query_analyst_store runs one read-only SQLite query over the same store the InfraMap agent uses. It is the fastest way to screen, rank, count or compare across every mapped project. Join tables on project_id.

projects

One row per grouped project: name, operator, location, primary fuel, nameplate MW by technology, battery MW and MWh, ISO, balancing authority, current price node, comparison hub and the reviewed price-node status.

generation_annual

EIA-923 net generation by year from 2000, battery charge and discharge, months reported and capacity factor (null unless all 12 months of generation and capacity are present).

generation_monthly

The same by month from January 2019.

ownership

Reported and ultimate owners, and the reviewed offtake conclusion with its as-of date.

offtake_contracts

FERC market-based-rate and EQR contract records by buyer and seller. Rows can overlap through renewals and amendments.

prices_annual

2024 and 2025 time-weighted mean node and hub prices, basis and negative-price hours, day-ahead and real-time.

prices_annual_segments

Which node priced which part of a year when a project is priced through a price chain.

price_node_segments

Reviewed price chains: earlier pricing nodes of a current node across renames and market transitions.

hubs

Hub IDs and labels by ISO.

SELECT p.project_id, p.name, p.state, p.wind_mw,
       g.capacity_factor_pct
FROM projects p
JOIN generation_annual g ON g.project_id = p.project_id
WHERE p.iso = 'SPP' AND p.primary_fuel = 'Wind'
  AND g.year = 2025 AND g.complete_year = 1
ORDER BY p.wind_mw DESC
LIMIT 20

One SELECT or WITH statement of at most 6,000 characters. Results return at most 200 rows, and a query may scan at most 100,000 rows and run for at most 8 seconds. Recursive queries, ATTACH and value-generating functions are refused. Every result carries the store version and a citation for each table it reads.

What a result contains

Each tool returns structured content and the same JSON as text, for clients that read only text. Results keep the source’s own meaning:

• Evidence with dates, units, citations, coverage and exclusions. Null means unavailable, never zero. Zero and negative values are kept as reported.
• Native identities. Grouped projects keep their component plant and generator IDs, so nothing is counted twice.
• Pagination with exact offsets where a list is cut.
• A delivery envelope naming the server release, the data release and the analyst store version. Each dataset’s own observation and publication dates still apply.

Partial evidence stays partial. Modeled potential is not delivered generation, and modeled curtailment is an experimental estimate, not operator-reported curtailment. The Methodology page explains each model.

Limits

Request size

At most 32 KiB. One JSON-RPC message per request; batches are refused.

Result size

At most 160,000 bytes per tool result. Tools also apply their own row and period bounds.

Time

30 seconds per tool call.

Rate

About 120 requests per minute and 4 simultaneous requests per IP address.

Analyst SQL

200 rows returned, 100,000 rows scanned, 8 seconds.

Cold start

The first analyst-store query on a new server instance loads the store and takes a few seconds.

Browsers

Native clients need no Origin header. Requests from web pages are accepted only from allowed origins.

These are the server’s current limits, not a guarantee of capacity or availability. The server logs tool names, outcomes, latency and output size. It does not log request arguments or returned evidence.

What is not included

• The separate InfraMap data-center atlas.
• Third-party industry research. The server serves InfraMap’s reviewed datasets only.
• Unpublished research, ingestion operations, arbitrary web or file fetching, and any write access.
• Some price-node review evidence is internal. Results that rely on it say so and withhold it.

The full list of datasets the server reads is on Data sources.

Releases

The server is rebuilt from InfraMap Pro after each data release. The current release, data release and analyst store dates are in service.json(opens in a new tab).