Skip to main content
Glama
README.md
# dof-mcp

[![CI](https://github.com/amatiasdev/dof-mcp/actions/workflows/ci.yml/badge.svg?branch=main)](https://github.com/amatiasdev/dof-mcp/actions/workflows/ci.yml?query=branch%3Amain)
[![Contract](https://github.com/amatiasdev/dof-mcp/actions/workflows/contract.yml/badge.svg?branch=main)](https://github.com/amatiasdev/dof-mcp/actions/workflows/contract.yml?query=branch%3Amain)
[![OpenSSF Scorecard](https://api.scorecard.dev/projects/github.com/amatiasdev/dof-mcp/badge)](https://scorecard.dev/viewer/?uri=github.com/amatiasdev/dof-mcp)
[![OpenSSF Best Practices](https://www.bestpractices.dev/projects/15638/badge)](https://www.bestpractices.dev/projects/15638)
[![npm](https://img.shields.io/npm/v/dof-mcp)](https://www.npmjs.com/package/dof-mcp)

MCP server and typed TypeScript client for Mexico's **Diario Oficial de la Federación** (DOF), built on the public SIDOF JSON API, so AI agents (Claude Code, Claude Desktop, Cursor, etc.) can read official publications and financial indicators.

## Why

The DOF is where federal laws, decrees, tax and customs rules, and official indicators are published. Its JSON API is public, with only sparse official documentation (https://sidof.segob.gob.mx/datos_abiertos), and there is no typed client or MCP server for it on npm. This project wraps it with validation, provenance and a stable, normalized shape.

## MCP server

Three read-only tools:

| Tool | What it returns |
|---|---|
| `dof_daily_summary` | A day's publications with counts by edition and organization; filter by edition or organization |
| `dof_get_note` | One publication as plain text, paged with `offset` / `maxChars` for long notes |
| `dof_get_indicators` | FIX exchange rate, UDI and TIIE rates for a day, or a daily series of one indicator |

Every result includes:

- `officialUrl`: the publication on dof.gob.mx, so a person can check the official text
- `provenance`: the SIDOF URL the data came from, when it was fetched (UTC) and a SHA-256 of the raw response

### Install

Requires Node.js 22 or later. The server runs over stdio; add it to your MCP client:

```json
{
  "mcpServers": {
    "dof": { "command": "npx", "args": ["-y", "dof-mcp"] }
  }
}
```

- **Claude Desktop:** Settings → Developer → Edit Config, which opens `claude_desktop_config.json` (`%APPDATA%\Claude\` on Windows, `~/Library/Application Support/Claude/` on macOS).
- **Cursor:** `.cursor/mcp.json` in your project, or `~/.cursor/mcp.json` for every project.
- **Claude Code:** `claude mcp add dof -- npx -y dof-mcp`
- **Windows:** if your client cannot start `npx` directly, use `"command": "cmd", "args": ["/c", "npx", "-y", "dof-mcp"]`.

### Hosted endpoint (no install)

A public instance runs at **`https://mcp.aldomatias.com/mcp`** (Streamable HTTP, no authentication, rate limited). Use it from clients that accept remote MCP servers, such as custom connectors in Claude, or:

```bash
claude mcp add --transport http dof https://mcp.aldomatias.com/mcp
```

It serves the same tools and the same release as the npm package. It is a free, best-effort service with no SLA; for heavy use, run the package yourself.

**Privacy.** The endpoint does not store IP addresses, tool arguments or results. For each request it logs the HTTP method and path, status, duration, the JSON-RPC method, the tool name, and the name and version your MCP client reports about itself (for example `claude-ai`). Those logs are kept for 7 days in AWS CloudWatch and are used only to count usage and detect failures. Traffic passes through Cloudflare, which processes IP addresses to apply rate limiting and DDoS protection under [its own privacy policy](https://www.cloudflare.com/privacypolicy/). Questions: contacto@aldomatias.com.

### Streamable HTTP

```bash
npx dof-mcp --http --port 3000   # endpoint http://127.0.0.1:3000/mcp, probe /healthz
```

The HTTP transport is stateless, binds to `127.0.0.1` and validates the `Host` and `Origin` headers (`--allowed-host <name>` adds hostnames). It has no authentication or rate limiting: to expose it beyond localhost, put it behind a reverse proxy that adds them. See [SECURITY.md](SECURITY.md).

## Client

The package is ESM. `import` works everywhere; `require("dof-mcp")` works on Node.js 22.12 and later (`require(esm)`).

```ts
import { DofClient } from "dof-mcp";

const dof = new DofClient();

const notes = await dof.getNotes("2026-10-16");
// [{ id: 5801377, title: "Acuerdo por el que se crea ...", date: "2026-10-16",
//    edition: "matutina", branch: "PODER EJECUTIVO",
//    organization: "SECRETARIA DE SEGURIDAD Y PROTECCION CIUDADANA", page: 4,
//    hasPdf: true, hasHtml: true,
//    officialUrl: "https://dof.gob.mx/nota_detalle.php?codigo=5801377&fecha=16/10/2026" }, ...]

const note = await dof.getNote(5801457); // full text as plain text
const fix = await dof.getIndicatorSeries("fix", "2026-10-01", "2026-10-16");

const { data, source } = await dof.getNotesWithSource("2026-10-16");
// source: { sourceUrl, fetchedAt, sha256, bytes }
```

Options (all optional):

| Option | Default | |
|---|---|---|
| `timeoutMs` | `15000` | Per-request timeout |
| `retries` | `2` | Retries for network errors, 429 and 5xx, with exponential backoff and full jitter; honors `Retry-After` up to 30 s |
| `maxResponseBytes` | 10 MiB | Larger responses are rejected unparsed |
| `cacheTtlMs` | 10 min | In-memory cache; identical concurrent requests share one fetch. `0` disables |
| `cacheMaxEntries` | `256` | LRU bound |
| `maxCacheableBytes` | 512 KiB | Larger responses are returned but not cached |
| `maxConcurrency` | `4` | Requests in flight to SIDOF at once |
| `maxQueue` | `64` | Requests waiting for a slot; beyond it they fail fast with `DofQueueFullError` |

## Design notes

- **Validation at the boundary.** Every response is parsed with Zod; an unexpected shape fails loudly instead of leaking `undefined` into callers.
- **Normalization.** SIDOF mixes section headers and publications in one list and writes some missing values as the string `"null"`; the client returns titled publications only, with ISO dates, booleans and plain-text content.
- **Polite to the source.** Identifying `User-Agent`, concurrency limit, cache, no redirects. SIDOF answers `406` to a bare `Accept: application/json`, so the client sends a header with a fallback.
- **Recorded fixtures.** Tests run against real recorded responses; a nightly contract test checks the live API still matches.

## Roadmap

- [x] Typed client for daily publications, note detail and indicators (FIX, UDIS, TIIE)
- [x] MCP server (stdio + Streamable HTTP) with `dof_daily_summary`, `dof_get_note`, `dof_get_indicators`
- [x] Hardening: size limits, backoff with jitter, cache, provenance; nightly contract test
- [x] Release pipeline: npm trusted publishing with provenance, MCP Registry
- [ ] Optional: hosted endpoint, local search index (`dof_search_notes`; SIDOF's own search is broken)

## Development

See [CONTRIBUTING.md](CONTRIBUTING.md) and [RELEASING.md](RELEASING.md).

```bash
npm install
npm run lint && npm run typecheck && npm test && npm run build
npm run test:live   # contract tests against the live API
```

## Disclaimer

Unofficial project, not affiliated with the Secretaría de Gobernación. Data comes from the public SIDOF API and may change without notice. The text this package returns is a plain-text conversion, not the official edition: the official DOF is the signed PDF on dof.gob.mx, so always confirm against it.

The MIT license covers this code, not DOF content. SIDOF's [legal notice](https://sidof.segob.gob.mx/avisoLegal) allows reproducing its content for non-commercial purposes when the source is cited; commercial use requires written consent from the Secretaría de Gobernación. If you republish content obtained with this package, cite the Diario Oficial de la Federación as the source.

## License

MIT

TDQS

A4/5.0

Scored across 3 tools

Disambiguation5/5

Each tool targets a distinct object and action: a daily index of publications, a single publication's content, and economic indicators. The descriptions explicitly chain them ('Use the returned ids with dof_get_note') and clarify that indicators are a separate query mode, leaving no realistic misselection risk.

Naming Consistency4/5

All three share a clear 'dof_' namespace, but two use the verb_get pattern (dof_get_note, dof_get_indicators) while the third is a noun phrase (dof_daily_summary). The deviation is minor and readable.

Tool Count4/5

Three tools is at the thin end of a well-scoped set, but each covers a genuinely different need (index, document, series data) with no filler. It is slightly under-powered for a document-retrieval domain rather than over-scoped.

Completeness3/5

The core lifecycle (list a day, fetch a note with pagination, pull indicator series) is covered, but there is no way to search publications by keyword, organization, or date range across days, forcing agents to crawl day by day. That is a notable gap for a lookup service.

Maintenance

ActivityMaintained
ResponsivenessLimited data