trailsearch-mcp
# trailsearch-mcp
MCP server dla self-hostowanego **TrailSearch / open-tavily**
(jianjungki/tavily-open) — API kompatybilne z Tavily, oparte o SearXNG.
Gotowe MCP (`tavily-ai/tavily-mcp`, `kshern/mcp-tavily`) hardkodują
`https://api.tavily.com` i nie obsługują własnego base URL — ten pakiet
wskazuje na Twoją instancję.
## Narzędzia
| Narzędzie MCP | Endpoint open-tavily |
|---|---|
| `trailsearch_search` | `POST /tavily/search` |
| `trailsearch_extract` | `POST /extract` |
## Uruchomienie (self-healing, jak `fetch`/`searxng`)
Pakiet jest instalowany **na żądanie** z GitHub przez `uvx` — żaden lokalny
plik nie musi przetrwać:
```bash
TAVILY_BASE_URL=http://192.168.1.134:3000 \
uvx --from git+https://github.com/marcinkuk/trailsearch-mcp.git trailsearch-mcp
```
## Konfiguracja w agent-canvas (mcp_config)
```json
"trailsearch": {
"command": "uvx",
"args": [
"--from",
"git+https://github.com/marcinkuk/trailsearch-mcp.git",
"trailsearch-mcp"
],
"env": { "TAVILY_BASE_URL": "http://192.168.1.134:3000" },
"enabled": true
}
```
Po zmianie/nowym starcie konwersacji agent dostaje `trailsearch_search` /
`trailsearch_extract`.
Wymagania: `uvx` (uv) w obrazie agenta oraz internet (git clone na żądanie).
TDQS
Scored across 2 tools
The two tools have clearly distinct purposes: one performs web searches, the other extracts content from known URLs. There is no overlap or ambiguity between them.
Both tools follow a consistent 'trailsearch_' prefix followed by a verb in lowercase snake_case, which is used uniformly. This matches a clear and predictable pattern.
With only two tools, the server feels minimal, though it covers the essential search and extraction workflow. The low count is borderline and fits the 'thin' end of the scale.
The pair provides solid coverage of search and content extraction, with search supporting many filters and answer generation. Minor gaps exist (e.g., no dedicated endpoint for querying recent trends or managing custom indices), but core user needs are met.