freecase-mcp
# freecase-mcp
An [MCP](https://modelcontextprotocol.io) connector that lets AI agents —
Claude Desktop, Cursor, Hermes, and any other MCP host — search and read the
[Freecase](https://freecase.ai) legal corpus: U.S. case law (Illinois, federal,
the Supreme Court, the 7th Circuit) and a growing set of state and federal
statutes.
It is a thin client. It calls the public Freecase backend over HTTPS and never
touches a database — nothing to install a database driver for, no credentials on
your machine beyond your personal connector key.
---
## What you get
Five tools:
| Tool | Needs a key? | What it does |
|------|:---:|------|
| `search_cases` | Yes | Full-text case-law search (Illinois, federal, SCOTUS, 7th Circuit). Returns ranked results with citations, courts, dates, and snippets. |
| `search_statutes` | Yes | Statute search by citation (`625 ILCS 5/11-501`) or plain text, across the launched jurisdictions. |
| `get_opinion` | No | Full text of one opinion by `cluster_id`, plus its cited-by list and parentheticals. |
| `get_statute` | No | Full text of one statute section by `section_id`, including repealed/renumbered history. |
| `get_citations` | No | The parsed reporter citations for one case. |
The two `search_*` tools require a **connector key**. The three `get_*` tools
are public and work with no key at all.
---
## Prerequisites
You need a way to run the connector on your machine. The config examples below
use [`uv`](https://docs.astral.sh/uv/) (specifically `uvx`), which downloads and
runs `freecase-mcp` in one step with no manual install.
Install `uv` once:
- **macOS / Linux:** `curl -LsSf https://astral.sh/uv/install.sh | sh`
- **Windows (PowerShell):** `powershell -c "irm https://astral.sh/uv/install.ps1 | iex"`
`uv` bundles its own Python, so you do not need a separate Python install.
> **Note:** `uvx freecase-mcp` works once the package is published to PyPI.
> Until then, run it from a local checkout — see
> [Running from a local checkout](#running-from-a-local-checkout).
---
## Get your connector key
1. Sign in at [freecase.ai](https://freecase.ai) and open your **account page**.
2. In the **Connector keys** section, click **Generate key**.
3. Copy the key — it is shown **once** and starts with `fc_mcp_`.
The account page can also generate a ready-to-paste config block for each
platform below, with your key already filled in. If you use that, you can skip
straight to the restart step.
> **Protect your key like a password.** Anyone who has it can run searches
> billed to your account. Don't commit it to git, don't paste it into a shared
> doc, and don't sync the config file to a public location. If a machine or
> config is ever compromised, revoke the key on your account page and generate a
> new one — revocation is immediate.
---
## Set it up in your agent
Each MCP host has its own config file. Add a `freecase` server entry, put your
key in the `env` block, then **restart the app** — MCP hosts only load new
servers on restart, and a skipped restart is the most common "it isn't working"
cause.
### Claude Desktop
Edit `claude_desktop_config.json`:
- **macOS:** `~/Library/Application Support/Claude/claude_desktop_config.json`
- **Windows:** `%APPDATA%\Claude\claude_desktop_config.json`
```json
{
"mcpServers": {
"freecase": {
"command": "uvx",
"args": ["freecase-mcp"],
"env": {
"FREECASE_MCP_KEY": "fc_mcp_your_key_here"
}
}
}
}
```
Then quit and reopen Claude Desktop.
### Cursor
Edit `~/.cursor/mcp.json` (global) or `.cursor/mcp.json` in your project:
```json
{
"mcpServers": {
"freecase": {
"command": "uvx",
"args": ["freecase-mcp"],
"env": {
"FREECASE_MCP_KEY": "fc_mcp_your_key_here"
}
}
}
}
```
Then restart Cursor (or toggle the server off/on in Settings → MCP).
### Hermes
Add the server to your Hermes MCP config (`mcpServers` block):
```json
{
"mcpServers": {
"freecase": {
"command": "uvx",
"args": ["freecase-mcp"],
"env": {
"FREECASE_MCP_KEY": "fc_mcp_your_key_here"
}
}
}
}
```
Then restart Hermes so it picks up the new server.
### Pointing at a non-production backend
To talk to a local or staging backend, add `FREECASE_API_BASE` to the `env`
block (it defaults to `https://api.freecase.ai`):
```json
"env": {
"FREECASE_MCP_KEY": "fc_mcp_your_key_here",
"FREECASE_API_BASE": "http://localhost:8000"
}
```
---
## Try it
Once configured and restarted, ask your agent something like:
- "Search Illinois case law for proximate cause in a slip-and-fall."
- "Look up 410 U.S. 113 and summarize the holding."
- "Find the Illinois DUI statute 625 ILCS 5/11-501 and show me the text."
- "What cases cite cluster 12345?"
The agent will call `search_cases` / `search_statutes` / `get_opinion` and read
the results back to you.
---
## Running from a local checkout
Until the package is on PyPI (or for development), point the host at a checkout
instead of `uvx`:
```json
{
"mcpServers": {
"freecase": {
"command": "/absolute/path/to/freecase-mcp/.venv/bin/freecase-mcp",
"env": {
"FREECASE_MCP_KEY": "fc_mcp_your_key_here"
}
}
}
}
```
Or with `uv` from the project directory:
```json
{
"mcpServers": {
"freecase": {
"command": "uv",
"args": ["run", "--directory", "/absolute/path/to/freecase-mcp", "freecase-mcp"],
"env": { "FREECASE_MCP_KEY": "fc_mcp_your_key_here" }
}
}
}
```
You can also run `python -m freecase_mcp` from an environment where the package
is installed.
---
## Configuration reference
| Env var | Required | Default | Notes |
|---------|:---:|---------|-------|
| `FREECASE_MCP_KEY` | For search only | *(none)* | Your connector key (`fc_mcp_...`). The `get_*` tools work without it. |
| `FREECASE_API_BASE` | No | `https://api.freecase.ai` | Override only for local/staging backends. |
Both are read at call time, so a host that injects env vars late still works.
---
## Development
```bash
python3 -m venv .venv
. .venv/bin/activate
pip install -e ".[dev]"
pytest -q # unit tests: fully mocked, no network, no database
```
`scripts/smoke.py` runs a handful of real queries end to end against a live
backend. It is **not** part of the automated test suite — it needs a real
`FREECASE_MCP_KEY` and a reachable backend. Run `python scripts/smoke.py
--help` for details.
---
## Not yet available
These are planned but **not implemented** in this version. Silence here means
"not built yet," not "not planned":
- **`check_treatment`** — citator treatment (overruled / questioned / etc.) for a case.
- **`find_cases_citing_statute`** — cases that cite a given statute section.
- **`search_by_citation`** — dedicated citation-lookup tool (for now, just pass a
citation string to `search_cases` — Freecase detects citations server-side).
- **`get_court_hierarchy`** — court structure / binding-authority relationships.
- **`get_parentheticals`** — a standalone parentheticals tool (parentheticals are
currently returned as part of `get_opinion`, not on their own).
- **SSE / remote hosted deployment** — this version is **stdio-local only**: it
runs on your machine and speaks to the backend over HTTPS. A hosted remote MCP
endpoint (one URL, no local install) is a possible future step, not a current
feature.
---
## How it fits together
```
Your agent ──stdio──▶ freecase-mcp (this package) ──HTTPS──▶ api.freecase.ai
```
The connector holds no legal data and no database credentials. All search,
ranking, and query handling happen on the Freecase backend; this package just
forwards requests (adding your connector key on the two search tools) and
returns the results.
TDQS
Scored across 5 tools
Each tool has a distinct purpose: search cases vs. search statutes, and retrieval for citations, full opinion, and statute text. There is no overlap between the functions, and the descriptions clearly delineate their respective domains.
All tools follow a consistent verb_noun pattern with snake_case: 'get_citations', 'get_opinion', 'get_statute', 'search_cases', 'search_statutes'. The convention is uniform and predictable.
5 tools is well-scoped for a legal research server covering both case law and statutes. Each tool serves a necessary role without unnecessary bloat or missing core functionality.
The tool set covers search and retrieval for cases and statutes comprehensively. Minor gaps exist, such as no tool for listing jurisdictions or code families, but the core CRUD-like workflow (search, get details) is fully supported.