solar-mcp
Open-source MCP servers exposing US solar data (production, resource, economics, market, forecast) as agent-accessible tools with a consistent data envelope.
README
<div align="center">
☀️ solar-data-mcp
US solar data, agent-accessible.
One MCP server that brings US open solar data to Claude, ChatGPT, and anything else that speaks MCP.
</div>
Install one server — uvx solar-data-mcp — and your agent gets all 18 tools, 11
skills that teach it how to use them, and 4 ready-made report prompts, across four
data domains: production modeling, economics, market data, and forecasts. Every tool
returns the same envelope — data + units + source + assumptions + warnings —
so the agent always knows what a number means, where it came from, and which defaults
were injected on its behalf.
What can you ask it?
- Thinking about solar at home — "Would a 6 kW system pay off at my house?" · "Is this $21,000 quote fair?" · "What will my array generate tomorrow?"
- Selling or installing solar — "Build a proposal for this customer" · "How long does permitting take in Phoenix?" · "Where should we expand next?"
- Studying the market — "Brief me on the Texas solar market" · "How have installed prices trended in Colorado?" · "Which big plants have batteries?"
Behind each question the agent picks the right tools (or a skill routes it), and every number comes back with units, a source, and the assumptions made on your behalf.
Quickstart
-
Get a free NREL API key — https://developer.nlr.gov/signup/ (or use
DEMO_KEYto try it out — 10 requests/hour). -
Add the server to your agent — snippets for every major agent below; for Claude Desktop, merge this into
claude_desktop_config.json(full example inexamples/):{ "mcpServers": { "solar-data": { "command": "uvx", "args": ["solar-data-mcp"], "env": { "NREL_API_KEY": "YOUR_KEY_HERE", "OPENEI_API_KEY": "YOUR_KEY_HERE", "EIA_API_KEY": "YOUR_KEY_HERE" } } } }Only
NREL_API_KEYis needed to start: the server runs with any subset of keys, and a tool missing its key returns setup instructions instead of failing silently. -
Restart your client and ask — try "Compare annual production for an 8 kW system in Mesa, AZ at 10° vs 25° tilt."
Verify keys and connectivity anytime:
$ uvx solar-data-mcp doctor
Add it to your agent
Claude Code
$ claude mcp add solar-data \
--env NREL_API_KEY=YOUR_KEY --env OPENEI_API_KEY=YOUR_KEY --env EIA_API_KEY=YOUR_KEY \
-- uvx solar-data-mcp
(or commit the quickstart JSON to your project's .mcp.json)
Codex CLI — ~/.codex/config.toml:
[mcp_servers.solar-data]
command = "uvx"
args = ["solar-data-mcp"]
[mcp_servers.solar-data.env]
NREL_API_KEY = "YOUR_KEY_HERE"
OPENEI_API_KEY = "YOUR_KEY_HERE"
EIA_API_KEY = "YOUR_KEY_HERE"
(or codex mcp add solar-data --env NREL_API_KEY=YOUR_KEY -- uvx solar-data-mcp;
use env_vars = ["NREL_API_KEY"] to forward keys from your shell instead of
hardcoding them)
OpenCode — opencode.json in your project (or ~/.config/opencode/opencode.json):
{
"$schema": "https://opencode.ai/config.json",
"mcp": {
"solar-data": {
"type": "local",
"command": ["uvx", "solar-data-mcp"],
"enabled": true,
"environment": {
"NREL_API_KEY": "YOUR_KEY_HERE",
"OPENEI_API_KEY": "YOUR_KEY_HERE",
"EIA_API_KEY": "YOUR_KEY_HERE"
}
}
}
}
Hermes — ~/.hermes/config.yaml, then /reload-mcp:
mcp_servers:
solar-data:
command: "uvx"
args: ["solar-data-mcp"]
env:
NREL_API_KEY: "YOUR_KEY_HERE"
OPENEI_API_KEY: "YOUR_KEY_HERE"
EIA_API_KEY: "YOUR_KEY_HERE"
Claude Desktop — the quickstart JSON above.
Anything else that speaks MCP (stdio) — command uvx, args ["solar-data-mcp"],
keys in the env block.
API keys
All free, and every one optional — the server starts with none set.
| Env var | Unlocks | Get one |
|---|---|---|
NREL_API_KEY |
estimate_production, get_solar_resource, compare_orientations, size_system_for_target, estimate_roi, compare_forecast_to_model | https://developer.nlr.gov/signup/ |
OPENEI_API_KEY |
lookup_tariffs | https://openei.org/services/api/signup/ |
EIA_API_KEY |
get_electricity_prices | https://www.eia.gov/opendata/register.php |
AHJ_REGISTRY_TOKEN (optional) |
identify_ahj | email support@sunspec.org |
Market tools (USPVDB, Tracking the Sun, SolarTRACE) and forecasts need no key.
Full forecast output additionally needs the Quartz model installed into a persistent
environment (see packages/solar-forecast/);
without it the forecast tools return install instructions.
Servers
uvx solar-data-mcp — the install above — serves all four domains on one stdio entry.
Each domain also ships as a standalone server:
| Domain | Data | Standalone server |
|---|---|---|
| Production | PVWatts v8 modeling, NSRDB irradiance | uvx --from solar-data-mcp-nrel nrel-solar-mcp |
| Economics | URDB tariffs, EIA prices, federal ITC + DSIRE | uvx --from solar-data-mcp-economics solar-economics-mcp |
| Market | SolarTRACE, Tracking the Sun, USPVDB, AHJ lookup | uvx --from solar-data-mcp-market solar-market-mcp |
| Forecast | Quartz open-source forecasts (OCF) | uvx --from solar-data-mcp-forecast solar-forecast-mcp |
Per-server config:
examples/claude_desktop_config.per-server.json.
⚠️ Run the combined
solar-dataserver or the per-domain servers, not both — and neversolar-economicsandsolar-marketside by side. Both open the same local DuckDB bulk store, which allows only one process at a time.
Tools
Eighteen tools across four domains; parameter details live in each tool's docstring and each domain package's README.
| Domain | Ask it about | Tools |
|---|---|---|
| Production | output, sunniness, sizing, roof orientation | estimate_production, get_solar_resource, compare_orientations, size_system_for_target |
| Economics | tariffs, electricity prices, incentives, payback | lookup_tariffs, get_electricity_prices, get_incentives, estimate_roi, sync_incentives |
| Market | installed $/W, permitting times, utility-scale plants | query_installed_systems, get_permitting_timelines, find_utility_scale_projects, identify_ahj, market_snapshot, sync_* |
| Forecast | tomorrow's output, "is today unusual?" | forecast_generation, compare_forecast_to_model |
Skills & reports
Skills are procedures shipped inside the combined server that teach an agent to chain
the tools correctly — ordering, sync prerequisites, honest reporting. They're MCP
resources: skill://solar/index routes by question shape, skill://solar/<name> is
the procedure. Grouped by who's asking:
- Homeowners — site assessment, quote review, performance check
- Installers — proposal builder, territory expansion
- Analysts — market brief, pricing analysis, utility-scale scout, incentive scan
- Cross-cutting — data sync (bulk snapshots), data conventions (envelope literacy)
Four of these render reports with a fixed document shape and are also exposed as
MCP prompts your host surfaces natively — market_brief, site_assessment,
quote_review, proposal_builder (in Claude Code: /mcp__solar-data__market_brief).
Full catalog, routing design, and report templates: docs/skills.md.
Development
$ git clone https://github.com/hoodsy/solar-data-mcp && cd solar-data-mcp
$ uv sync # install the workspace
$ uv run pytest # fixture replay only, no network
Layout: packages/core (shared HTTP client, cache, envelope), one package per domain
server, and packages/solar-data-mcp (the umbrella that mounts all four on one stdio
entry). Smallest possible client:
examples/example_client.py.
License
MIT. Per-source data licensing/attribution is exposed as MCP resources
(source://<name>/license).
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。