Metadata-Version: 2.5
Name: ecfr-mcp
Version: 1.0.9
Summary: MCP server for the eCFR (Electronic Code of Federal Regulations) API. Read FAR, DFARS, and all agency FAR supplements.
Project-URL: Homepage, https://1102tools.com
Project-URL: Repository, https://github.com/1102tools-dev/federal-contracting-mcps
Project-URL: Issues, https://github.com/1102tools-dev/federal-contracting-mcps/issues
Author: James Jenrette / 1102tools
License: MIT
Keywords: 1102,cfr,dfars,ecfr,far,federal-acquisition,mcp,model-context-protocol,regulations
Classifier: Development Status :: 5 - Production/Stable
Classifier: Intended Audience :: Developers
Classifier: License :: OSI Approved :: MIT License
Classifier: Programming Language :: Python :: 3
Classifier: Programming Language :: Python :: 3.10
Classifier: Programming Language :: Python :: 3.11
Classifier: Programming Language :: Python :: 3.12
Classifier: Programming Language :: Python :: 3.13
Classifier: Programming Language :: Python :: 3.14
Classifier: Topic :: Software Development :: Libraries :: Python Modules
Requires-Python: >=3.10
Requires-Dist: filelock>=3.13
Requires-Dist: httpx>=0.27.0
Requires-Dist: mcp<3,>=2.0.0
Requires-Dist: platformdirs>=4.0
Description-Content-Type: text/markdown

# ecfr-mcp

<!-- mcp-name: com.1102tools/ecfr-mcp -->

MCP server for the eCFR (Electronic Code of Federal Regulations) API. Read FAR, DFARS, and all agency FAR supplement text with no authentication required.

Connect through the published ChatGPT plugin below, or use the installation and configuration instructions for another compatible MCP client.

*Tested and hardened through six rounds of integration testing against the live eCFR API. 295 regression tests (182 offline, 113 live-gated) covering 2 P0 catastrophic bugs, 26 P1 silent-wrong-data bugs, 32 P2 validation gaps, and the round-6 audit fixes (Title 48 chapter whitelist, table extraction, appendix access). See [testing.md](testing.md) for the full testing record.*

## Available in ChatGPT

[Install eCFR in ChatGPT](https://chatgpt.com/plugins/plugin_asdk_app_6a9ef0341b04819192935fd4e5cd9b34)

This MCP is also available as a published plugin in the ChatGPT directory. Open the listing to install and connect it; no user API key or local Python setup is required. Then try a [matching prompt](https://1102tools.com/#ecfr). Prompts that combine sources require every listed MCP to be connected.

The installation and configuration sections below cover direct setup in other compatible MCP clients.

## What it does

Exposes the eCFR API as 13 MCP tools covering regulatory text, structure, search, version history, and common acquisition workflows:

**Core endpoints**
- `get_latest_date` - Get the most recent available date for a CFR title (call before other tools)
- `get_cfr_content` - Get parsed regulatory text for a section, subpart, or part
- `get_cfr_structure` - Hierarchical table of contents
- `get_version_history` - Amendment history for a section or part
- `get_ancestry` - Breadcrumb hierarchy path
- `search_cfr` - Full-text search with hierarchy filters
- `list_agencies` - All agencies with their CFR references
- `get_corrections` - Editorial corrections for a title

**Workflow convenience**
- `lookup_far_clause` - One-call FAR/DFARS clause text lookup (auto-resolves date)
- `compare_versions` - Side-by-side text comparison at two dates
- `list_sections_in_part` - All sections in a FAR/DFARS part
- `find_far_definition` - Search FAR 2.101 for a term definition
- `find_recent_changes` - Sections modified since a given date

## No authentication required

The eCFR API is fully public. No API key, no registration, no auth headers. Just install and use.

## Installation

### Via uvx (recommended)

```bash
uvx ecfr-mcp
```

### Via pip

```bash
pip install ecfr-mcp
```

### From source

```bash
git clone https://github.com/1102tools-dev/federal-contracting-mcps.git
cd federal-contracting-mcps/servers/ecfr-mcp
pip install -e .
```

## Configuration

Use the configuration below as the server definition and adapt its placement to your compatible MCP client. For practical requests using this source, see the [prompt library](https://github.com/1102tools-dev/federal-contracting-prompts).

```json
{
  "mcpServers": {
    "ecfr": {
      "command": "uvx",
      "args": ["--refresh-package", "ecfr-mcp", "--from", "ecfr-mcp", "ecfr-mcp"]
    }
  }
}
```

The `--refresh-package` flag tells uv to check PyPI for a newer release each time your client launches the server, so fixes arrive automatically; without it, uv keeps serving whatever version it first cached. It adds a moment of network time at startup, so raise your platform's MCP startup timeout if it enforces a short one.

Restart the client. The `ecfr` server appears with 13 tools.

## Example prompts

- "Pull the current text of FAR 15.305 (Proposal Evaluation) and summarize what it requires."
- "List all sections in FAR Part 19 (Small Business Programs)."
- "Look up the FAR definition of 'commercial product' in 2.101."
- "What FAR sections were amended in the last 6 months?"
- "Compare FAR 52.212-4 between 2024-01-01 and 2025-01-01 and show me what changed."
- "Get the current text of DFARS 252.227-7014 (Rights in Noncommercial Computer Software)."
- "Search Title 48 for 'organizational conflict of interest' and show me the relevant sections."
- "Which agency owns Chapter 8 in Title 48? Get their FAR supplement structure."

## Design notes

- **XML parsed server-side.** The eCFR content endpoint returns raw XML. This server parses it into clean text (headings, paragraphs, citations) before returning to the model, saving significant context tokens.
- **Automatic date resolution.** eCFR lags 1-2 business days behind the Federal Register. Using today's date on versioner endpoints causes 404 errors. All content tools auto-resolve to the latest available date unless you specify one.
- **Search defaults to current text.** Without `date=current`, eCFR search returns ALL historical versions including superseded. Default `current_only=True` prevents duplicate results.
- **Structure endpoint limitation.** The eCFR structure endpoint does not support section-level filtering (returns 400). `list_sections_in_part` works around this by fetching the part structure and walking the tree.
- **FAR 2.101 optimization.** The definitions section is ~109KB of XML. `find_far_definition` parses the full section server-side and returns only matching paragraphs with context.

## CFR Title 48 quick reference

| Chapter | Regulation | Parts |
|---|---|---|
| 1 | FAR | 1-99 |
| 2 | DFARS | 200-299 |
| 3 | HHSAR | 300-399 |
| 4 | AGAR | 400-499 |
| 5 | GSAR | 500-599 |
| 6 | DOSAR | 600-699 |
| 7 | AIDAR | 700-799 |
| 8 | VAAR | 800-899 |
| 9 | DEAR | 900-999 |
| 18 | NFS | 1800-1899 |

## Data source

All data from [ecfr.gov](https://www.ecfr.gov), the continuously updated online Code of Federal Regulations maintained by the Office of the Federal Register. Updated daily, typically 1-2 business days after Federal Register publication. Not an official legal edition; for official citations reference the annual CFR from GPO.

## Part of

[federal-contracting-mcps](https://github.com/1102tools-dev/federal-contracting-mcps): monorepo of 9 MCP servers for federal contracting data. Pair these sources with the [MCP prompt library](https://github.com/1102tools-dev/federal-contracting-prompts).

## Request pacing and cache

JSON requests can start every 0.6 seconds, with a shared safety budget of
500 upstream attempts per rolling five minutes and at most two requests in
flight. Uncached XML remains serialized with at least three seconds after
completion. XML waiting does not hold up JSON work. These are 1102tools
safeguards, not published agency quotas or guaranteed tool-call throughput.

Repeated XML requests with the same date, path and filters use a five-minute
in-memory cache. Concurrent duplicate requests are coalesced. The cache stores
only valid XML, at most 128 entries and 32 MiB total, with a 2 MiB per-entry
limit. Errors are not cached. New dates and changed filters use separate keys;
upstream corrections may take up to five minutes to appear in a cached result.
Restarting the process clears the cache.

Local processes sharing `FEDERAL_API_PACING_DIR` share upstream pacing state.
The hosted service's budget is shared by its users; other services have their
own budgets. Its separate HTTP entrance limit is 120 requests per minute per
IP and Cloudflare location, including protocol traffic. A tool can make more
than one upstream request, or none when XML is cached.

`FEDERAL_API_MIN_INTERVAL_SECONDS` can slow requests. Positive values below
0.6 are clamped for the shared pacer; uncached XML always retains its minimum
three-second completion gap. Explicit zero disables pacing for offline tests
or externally managed clients. Hosted deployments use 0.6. Failed and cancelled
attempts stay counted; Retry-After cooldowns are shared. Container replacement
can reset on-disk pacing history. Upgrade local processes together rather than
sharing a state directory between old and new implementations.

## License

MIT
