ferric-fred-mcp
A strongly-typed, single-binary MCP server for FRED (Federal Reserve Economic Data), written in Rust.
README
ferric-fred
A strongly-typed Rust client for FRED — the Federal Reserve Economic Data service from the Federal Reserve Bank of St. Louis — plus a CLI (with TUI charts) and an MCP server built on top of it.
ferric(iron oxide → rust) +FRED. Iron-clad, typed access to economic data.
Workspace
A Cargo workspace of three crates — each with its own README (the crates.io / docs.rs landing page) that carries the full usage detail:
| Crate | Binary | What it is | Details |
|---|---|---|---|
ferric-fred |
— | Strongly-typed async FRED client | README · docs.rs |
ferric-fred-cli |
fred |
Command-line tool with ratatui TUI charts |
README |
ferric-fred-mcp |
fred-mcp |
MCP server exposing FRED to MCP clients | README |
Published versions (these badges are the source of truth — the crates version independently, so they can drift out of lockstep):
Consumers depend on the library by workspace path, so a breaking change in the library cannot compile-pass its consumers without updating them — that compile-time coupling is the primary "stay in sync" guarantee (versions are managed on top; see the ADRs).
What it covers
The library wraps all of FRED's read endpoints — series and observations
(including ALFRED point-in-time / vintage data via a real-time window),
search, categories, releases (including the nested release-table tree, with
optional inline observation values), sources,
and tags — plus the GeoFRED / Maps API (regional data and the geographic
shape files to map it, ADR-0025) — behind
ergonomic builders, with newtype identifiers, typed enums for
FRED's closed value sets, a non-panicking error taxonomy, and auto-pagination
(Paginate::send_all walks an endpoint to exhaustion, Paginate::stream yields
lazily; --all on the CLI). See ADR-0020 and
ADR-0021.
GeoFRED support spans the library, CLI (fred geofred), and MCP
(get_regional_data, get_series_data, get_series_group) layers. The one
exception is the geographic shapes/file endpoint, which is library/CLI-only —
a large projected-GeoJSON blob is poor ergonomics for an MCP tool caller
(ADR-0025).
Pick an entry point:
- Library —
cargo add ferric-fred; typed async access from your own code. See the crate README and docs.rs. - CLI (
fred) —cargo install ferric-fred-cli; search, show metadata, print or chart observations in the terminal, browse categories, releases, sources, and tags, and pull GeoFRED regional data and map shapes (fred geofred). See the crate README orfred <command> --help. - MCP server (
fred-mcp) —cargo install ferric-fred-mcp; 34 tools over stdio covering the same read surface, for MCP-capable clients (ADR-0010). Each tool declares input and output schemas plus behavioural annotations (ADR-0023). See the crate README.
The MCP server is listed and scored on Glama:
Development
A Nix flake provides a reproducible toolchain (nix develop, or direnv allow
once), but the project builds with a plain Rust toolchain too — Nix supplies the
environment, not the build (ADR-0008).
Contributor setup, the fmt/clippy/test gate, the tracked git hooks, and the
workflow for adding an endpoint live in CONTRIBUTING.md.
CI (ci.yml) runs that same offline gate on every push and PR; a dormant
live.yml runs the live FRED tests once an Infisical machine identity is
configured (ADR-0016).
Secrets
The client reads a free FRED API key from the FRED_API_KEY environment
variable (get one at https://fredaccount.stlouisfed.org/apikeys). Locally,
secrets are injected via Infisical + direnv
(ADR-0009):
cp .envrc.example .envrc # local, git-ignored entry point
infisical login # user auth (opens a browser)
infisical init # link this dir → project
direnv allow # load the shell + inject secrets on cd-in
Store the key with infisical secrets set FRED_API_KEY="…" --env=dev --path=/shared.
No Infisical? Just set it directly in your git-ignored .envrc:
export FRED_API_KEY="…" — the library only reads the env var and has no
dependency on Infisical.
Architecture decisions
Design decisions are recorded as ADRs in docs/adr/. Start with
the index.
License
Dual-licensed under MIT OR Apache-2.0, at your option — the Rust ecosystem
default (ADR-0006). See LICENSE-MIT
and LICENSE-APACHE. Unless you state otherwise, any
contribution you submit is licensed under the same dual terms (see
CONTRIBUTING.md).
This covers our code; FRED data itself is subject to the St. Louis Fed's terms of use, and you supply your own API key — the project ships no data and no key.
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。