trustcase-mcp
# Trustcase MCP server
A small, read-only MCP server that lets AI tools like Cursor and Claude search the Trustcase directory of practical AI agent use cases. Every result links back to https://trustcase.dev.
Published on npm as `trustcase-mcp`. Install with `npx -y trustcase-mcp`.
## Tools
| Tool | What it does |
|---|---|
| `search_use_cases` | Search by words, with optional `category`, `min_trust_score` and `limit` (max 25). Returns title, category, summary, trust score and the trustcase.dev link. |
| `get_use_case` | Full entry by `slug`: what the agent does alone, what waits for a human yes, trust score breakdown, setup prompt, tools, models, where it runs, and the link. |
Both tools are marked read-only. The server never writes anything anywhere.
## Data source
It reads the public feed at `https://trustcase.dev/api/use-cases.json` (cached for 1 hour). Point it somewhere else with `TRUSTCASE_FEED_URL`, which also accepts a local file path. `test/feed.json` is a snapshot built from the Trustcase repo (97 use cases).
## Build and test
```bash
npm install
npm run build
npm test # end to end over stdio and streamable HTTP against test/feed.json
```
## Use it in Cursor
Add to `~/.cursor/mcp.json` (or `.cursor/mcp.json` in a project):
```json
{
"mcpServers": {
"trustcase": {
"command": "node",
"args": ["/absolute/path/to/trustcase-mcp/dist/index.js"]
}
}
}
```
Once it is on npm this becomes `"command": "npx", "args": ["-y", "trustcase-mcp"]`.
## Use it in 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": {
"trustcase": {
"command": "node",
"args": ["/absolute/path/to/trustcase-mcp/dist/index.js"],
"env": { "TRUSTCASE_FEED_URL": "https://trustcase.dev/api/use-cases.json" }
}
}
}
```
Restart Claude Desktop, then ask something like "Find a Trustcase use case for invoice follow ups with a trust score over 80."
## Remote (streamable HTTP)
```bash
PORT=8787 node dist/index.js # or: node dist/index.js --http
```
The endpoint is `http://localhost:8787/mcp` (stateless, POST). A hosted copy could live at something like `https://mcp.trustcase.dev/mcp` later.
## Claude Code
```bash
claude mcp add trustcase -- node /absolute/path/to/trustcase-mcp/dist/index.js
```
TDQS
Scored across 2 tools
The two tools form a clear search-then-detail workflow: search_use_cases returns candidate slugs, and get_use_case retrieves the full entry for one slug. Their descriptions explicitly state this sequence, so an agent should not confuse them.
Both names follow the same snake_case verb_noun pattern: search_use_cases and get_use_case. The resource noun is consistent, making the pair predictable.
Two tools is minimal but appropriate for a read-only directory lookup service with a search/detail split. It is slightly thin compared to richer API surfaces, but no tool appears redundant.
The surface covers the core workflow: find a use case by task description and retrieve its full trust assessment, approval boundaries, setup prompt, and link. Minor gaps exist, such as browsing by category or fetching an entry by slug without first searching, but these are workable.