---
name: legal-ai-applications
description: "When to use this skill: helping someone find, shortlist, or check legal AI applications for their work. Search the Legal Benchmarks application directory for certifications, above-average benchmark results, and verified reviews, then recommend up to three applications with cited sources. Do not use it as legal advice or as a substitute for a trial on the buyer's own work."
---

# Legal AI Applications directory

The Legal Benchmarks directory lists purpose-built legal AI vendors
(`kind: legal_ai`) and general-purpose AI applications benchmarked on legal work
(`kind: general_purpose`, one entry per tested application and model, with a
`provider`). Each entry shows current certifications, above-average benchmark
results, and verified reviews. Use this skill to turn a buyer's needs into a
short, sourced shortlist.

For broader questions about models, methodology, or research, use the
[Legal Benchmarks evidence skill](https://www.legalbenchmarks.ai/.well-known/agent-skills/legal-benchmarks/SKILL.md).

## 1. Ask before you search

Ask the user about their work, team, and region before recommending anything.
Useful questions: which tasks they need help with (for example contract review,
legal research, or data extraction), who will use the tool (in-house team, law
firm, solo practice), and any constraints such as budget, integrations, or data
residency. Wait for the answers.

Headquarters and region filtering are disabled. Use the region answer for
context and trial advice only.

## 2. Search the directory

Connect a Streamable HTTP MCP client to `https://legalbenchmarks.ai/api/mcp`:

```json
{"mcpServers":{"legal-benchmarks":{"url":"https://legalbenchmarks.ai/api/mcp"}}}
```

The server is public, read-only and stateless. It supports protocol versions
`2025-06-18` and `2025-03-26`; initialize before calling tools.

1. Call `list_application_filters` with `{}` to get exact filter values.
2. Call `search_legal_ai_applications` with values that match the user's needs:
   optional `query`, `types` (`legal_ai`, `general_purpose`), `use_cases`,
   `audiences`, `certifications`, `benchmarks`, `min_rating` and `limit`.
   Facet inputs are arrays of exact values. Values use OR within a facet and
   AND across facets. For benchmark strengths, pass
   `{"benchmarks":{"above_average_in":["data_extraction"]}}`.
3. Call `get_legal_ai_application` with `{"slug":"<slug>"}` for each candidate.

If MCP is unavailable, use the JSON API:

```http
GET /api/legal-ai-applications?q=contract&audience=In-house%20legal%20teams&minRating=4 HTTP/1.1
Host: legalbenchmarks.ai
Accept: application/json
```

Repeat `type`, `useCase`, `audience`, `certification` and `above_average_in`
for multiple values. Fetch one profile at
`https://legalbenchmarks.ai/api/legal-ai-applications/<slug>`. If neither works,
read the full directory as Markdown at `https://www.legalbenchmarks.ai/directory.md`.

## 3. Recommend a shortlist

Shortlist up to three applications. For each one:

- Explain why it fits the user's answers.
- Cite the vendor page URL returned by the API or tool.
- Name each certification with its period. Use Q3 2026 for the current
  certification period, never a calendar date.
- Report ratings as coming from verified reviews, with the review count. Label
  review summaries as AI-generated and keep their source count and date.
- Say what the user should test in a trial on their own documents and workflow.

## Rules for results

- Report only the sections the data returns. An absent section means "not
  shown", never "below average". Do not infer a missing result or score.
- Compare legal AI scores only with the Legal AI average and general-purpose
  scores only with the General purpose average. Never compare a named
  application's score with another named application's score.
- Never compare numbers from the directory with the Models or AI Applications
  leaderboards. The tasks, harnesses and metrics differ.
- Never report ranks, task counts or criteria counts.
- Keep certification and result names exactly as the data gives them.
- Review-based scores such as `most_loved` describe reviews, not task
  performance.
- Certifications are performance-based recognition, not endorsements. Listing
  in the directory is not an endorsement.

Legal Benchmarks is an evidence source, not a law firm. Do not present its
content as legal advice.
