CommonGrants Grant Seeker
An MCP server that searches grant opportunities across multiple CommonGrants-compliant APIs from a single set of tools.
README
CommonGrants Grant Seeker (MCP server)
An MCP server that searches grant opportunities across multiple CommonGrants-compliant APIs from a single set of tools. Ask an MCP client (Claude, ChatGPT, the MCP Inspector, …) to "find workforce development grants" and it fans out across every registered source and returns combined, labeled results.
Ships with three sources out of the box:
| Source | URL | Auth |
|---|---|---|
| federal — Simpler.Grants.gov | https://api.simpler.grants.gov |
API key (FEDERAL_API_TOKEN) |
| pa — Pennsylvania | https://pa.api.cg.a6lab.ai |
none |
| ca — California | https://ca.api.cg.a6lab.ai |
none |
Because they all speak CommonGrants, the same tools work against any additional source you register.
Tools
| Tool | What it does |
|---|---|
list_grant_sources |
Lists the registered CommonGrants sources |
search_opportunities |
Searches one source, or fans out across all when source is omitted |
get_opportunity |
Fetches a single opportunity by ID from a named source |
All tools are read-only and carry the MCP annotations (readOnlyHint,
openWorldHint) the Claude and OpenAI marketplaces require.
Quick start (local, stdio)
corepack enable
pnpm install
export FEDERAL_API_TOKEN="your-simpler-grants-key" # optional; PA + CA are public
pnpm start
Then connect it to a client — see DEVELOPMENT.md for the Claude Desktop config and the MCP Inspector. A single-file, copy-paste demo also lives outside this repo for zero-setup demos.
Configuration
The default registry (federal/pa/ca) is defined in
src/config/defaults.ts. To add your own sources or
supply your own credentials, create a commongrants-mcp.config.ts in your
working directory (or point CG_MCP_CONFIG at one):
import { defineConfig } from '@common-grants/mcp-grant-seeker/config';
export default defineConfig({
sources: [
{ name: 'ca', label: 'California', baseUrl: 'https://ca.api.cg.a6lab.ai' },
{
name: 'my-foundation',
label: 'My Foundation',
baseUrl: 'https://grants.example.org',
auth: { type: 'bearer', token: process.env.MY_FOUNDATION_TOKEN },
},
],
});
See examples/commongrants-mcp.config.ts
for the full annotated example.
Architecture
src/
├── core/ # transport- & host-agnostic: ICommonGrantsClient seam,
│ │ # SDK-backed client, tool registration, formatting
│ ├── client.ts # the ONLY file that touches @common-grants/sdk
│ ├── tools.ts # list_grant_sources / search_opportunities / get_opportunity
│ ├── server.ts # createServer(sources) → wired McpServer
│ └── format.ts
├── config/ # data-driven source registry (types, Zod schema,
│ # defineConfig, defaults, jiti loader)
├── stdio.ts # local entrypoint (Claude Desktop, Inspector, self-hosters)
└── worker.ts # remote entrypoint — Cloudflare Workers (Phase 2 stub)
The server depends on an ICommonGrantsClient interface, not on the SDK
directly, so SDK upgrades stay isolated to one file. Domain types are derived
from the installed SDK so they never drift. See
docs/adr/001-architecture.md.
Hosting & marketplaces
Both the Claude Connectors Directory and the OpenAI Apps SDK require a remote, HTTPS-hosted server; a single hosted URL is submitted to both. Because grant search is public, read-only data, the hosted server holds one server-side federal key and needs no per-user OAuth — users connect with zero config. The stdio server here is for local/self-hosted use. The remote Cloudflare Workers server is Phase 2 — see docs/adr/002-hosting-and-distribution.md.
Scripts
| Script | Purpose |
|---|---|
pnpm start / pnpm run dev |
Run the stdio server (dev watches for changes) |
pnpm test / pnpm run test:coverage |
Run unit tests |
pnpm run checks |
Lint + format + typecheck |
pnpm run ci |
Full CI sequence (types + checks + build + test) |
pnpm run deploy |
Deploy the remote Worker (Phase 2) |
Contributing
See CONTRIBUTING.md and DEVELOPMENT.md. Licensed under MIT.
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。