Your AI agent can write code, summarize documents, and answer questions all day long. But ask it to draft a motion to compel or structure a legal research memo? It improvises. Missing sections. Inconsistent formatting. Work that doesn't meet the bar.
Today we're releasing Legal Agent Skills (Browse | Contribute), a curated library of professional-grade prompts and templates that give your agents domain expertise on demand.
The Problem We Kept Running Into
At case.dev, we build legal tech infrastructure. We also use AI coding assistants constantly. And we noticed something: these tools are brilliant at general software development but completely lost when it comes to legal-specific work.
They don't know what goes in a complaint. They've never seen a properly structured appellate brief. They improvise on legal research memos in ways that would make any attorney cringe.
So we built what we needed and made it available to everyone.
What's in the Box
Anthropic recently standardized a protocol for progressive disclosure called Agent Skills. Instead of cramming everything into a system prompt, agents load specialized knowledge only when they need it.
We took that approach and went deep on legal. Legal Agent Skills is a searchable library of battle-tested prompts, document templates, and research methodologies. Each skill includes:
- Structured prompts for generating specific legal documents
- Legal context metadata like jurisdiction, practice area, complexity level, matter phase
- Reference links connecting related skills and citation formats

Each skill carries metadata so your agent can route intelligently:
{
"slug": "legal-research-summaries",
"name": "Legal Research Summaries",
"legal_context": {
"complexity": "mid",
"jurisdictions": ["US"],
"practice_areas": ["litigation"],
"skill_modes": ["research", "summarization", "analysis"],
"matter_phases": ["pre-filing", "pre-trial"],
"document_types": ["summary", "analysis"]
}
}
Free MCP Access (No Strings)
Legal Agent Skills is available via Model Context Protocol with no authentication required:
| Property | Value |
|---|---|
| URL | https://skills.case.dev/api/mcp |
| Protocol | Streamable HTTP |
| Auth | None required (50 req/day free) |
Cursor
{
"mcpServers": {
"legal-skills": {
"url": "https://skills.case.dev/api/mcp"
}
}
}
Claude Desktop
{
"mcpServers": {
"legal-skills": {
"command": "npx",
"args": ["-y", "mcp-remote", "https://skills.case.dev/api/mcp"]
}
}
}
Windsurf
Navigate to Settings > MCP Servers, add a new server with name legal-skills and URL https://skills.case.dev/api/mcp.
Two Ways to Build
Option 1: Vibe Code It
If you're building legal tech and want to move fast, just connect the MCP server to your favorite AI coding tool and start building. Cursor, Claude Desktop, Windsurf, VS Code with Copilot... whatever you're using. The skills are there. Your agent will find them.
For Thurgood users, Agent Skills are already baked in. No setup required. Just describe what you want to build and Thurgood loads the relevant legal expertise automatically.
Option 2: Work With Us
If you're an enterprise or ALSP that needs custom implementations, integrations with your existing systems, or skills tailored to your specific practice areas and workflows... let's talk.
We've done this for court reporting companies, litigation support firms, and legal tech startups. We can help you integrate Legal Agent Skills into your existing infrastructure, build custom skills for your domain, or architect something entirely new.
Reach out at sales@casemark.com or grab time on my calendar. We're small, we're fast, and we actually pick up the phone.
The Tools Your Agent Gets
Once connected, your agent can use two tools:
resolve_skill
Search the library by topic, practice area, or document type.
"Find skills for drafting employment complaints"
"Search for appellate brief templates"
"What skills are available for corporate governance?"
Returns a ranked list of matching skills with name, summary, and relevance score.
read_skill
Retrieve the full content of a specific skill by slug.
"Read the legal-research-summaries skill"
"Get the judgment-summaries skill content"
Returns the complete skill including prompt content, legal context metadata, and references to related skills.
How Search Works
Skills uses hybrid search combining three methods:
- Text search for pattern matching on skill names and summaries
- Filter search matching on practice areas and document types extracted from your query
- Semantic search using vector similarity with embeddings
Results from all methods are merged and ranked. Skills appearing in multiple search methods get boosted.
REST API
For programmatic access beyond MCP:
# Search for skills
curl "https://api.case.dev/skills/resolve?q=appellate+brief"
# Read a skill
curl "https://api.case.dev/skills/legal-research-summaries"
Authenticated requests get unlimited access at $0.005 per request. Anonymous requests are limited to 50/day per IP.
Example: Building a Legal Research Assistant
async function handleLegalQuery(query: string) {
// 1. Find relevant skills
const skillsResponse = await fetch(
`https://api.case.dev/skills/resolve?q=${encodeURIComponent(query)}`
);
const { results: skills } = await skillsResponse.json();
if (skills.length === 0) {
return generateResponse(query, null);
}
// 2. Load the top skill
const skillResponse = await fetch(
`https://api.case.dev/skills/${skills[0].slug}`
);
const skill = await skillResponse.json();
// 3. Generate response using skill as guidance
return generateResponse(query, skill.content);
}
Pricing
| Tier | Cost | Limit |
|---|---|---|
| Free | $0 | 50 requests/day per IP |
| Authenticated | $0.005/request | Unlimited |
Get an API key at console.case.dev.
Get Started
- Add the MCP server to your AI tool (Cursor, Claude Desktop, Windsurf)
- Ask your agent to "search for legal research skills"
- Start building legal tech that actually knows legal
Or if you want to go deeper, reach out. We're here to help.
Documentation: docs.case.dev/agent-skills
MCP Endpoint: https://skills.case.dev/api/mcp
Enterprise (BAA, SSO, etc) & Custom Work: sales@casemark.com

