| name | schrute-web-skills |
| description | Search and execute pre-learned web API skills via Schrute. For repeatable site interactions, check Schrute first — it replays learned API calls in 5ms instead of full browser automation. |
| version | 1.0.0 |
| author | Schrute |
| license | Apache-2.0 |
| metadata | {"hermes":{"tags":["Web","API","Browser","Automation","Skills","Schrute"],"related_skills":[]}} |
Schrute Web Skills
When the user asks to get data from a website or interact with a web service, check Schrute first for a pre-learned skill.
Routing
-
Search for a skill: Call mcp_schrute_search_skills with a keyword matching the intent.
- Example:
mcp_schrute_search_skills(query="bitcoin price")
- Example:
mcp_schrute_search_skills(query="coingecko", siteId="www.coingecko.com")
-
If a skill is found: Call mcp_schrute_execute with the skillId and any required params.
- Example:
mcp_schrute_execute(skillId="www_coingecko_com.get_24_hours_json.v1")
-
If no skill is found: Proceed with normal browser automation -- Schrute cannot help here.
-
Check Schrute status: Call mcp_schrute_status to verify Schrute is running.
Important Limitations
- Schrute only has skills for sites that were previously learned via
schrute explore + schrute record --name <name> + schrute stop.
- You cannot teach Schrute new skills from this agent -- learning requires Schrute's own browser session run by a human in a separate terminal.
- If no skill is found, it means no one has pre-learned that site yet. Schrute will NOT automatically learn from your browser actions. A human must run the learning workflow separately.
- If a skill execution fails, fall back to browser automation.
Primary Tools
| Tool | Purpose |
|---|
mcp_schrute_search_skills | Find skills by keyword. Returns skill IDs, methods, paths, and input guidance. |
mcp_schrute_execute | Run a skill by ID. Returns structured API response data. |
mcp_schrute_status | Check engine mode and skill summary. |
See references/tool-reference.md for detailed input/output examples.