Documentation Index
Fetch the complete documentation index at: https://trygradient.ai/docs/llms.txt
Use this file to discover all available pages before exploring further.
Connectors
Connectors are data sources that candidates can browse and search during an assessment. They simulate the kind of enterprise tools and knowledge bases that professionals use daily — email, project trackers, meeting notes, internal wikis, and enterprise search.Connector categories
When adding connectors to an assessment, admins choose from categories. Each category has named variants so the connector appears branded to the candidate (e.g., “Gmail” instead of generic “Mail”).| Category | Variants | Description |
|---|---|---|
| Enterprise Search | Enterprise Search, Glean, Coveo | Company knowledge base with keyword search |
| Meeting Transcripts | Meeting Transcripts, Granola, Otter | Search meeting notes and call transcripts |
| Email, Gmail, Outlook | Search email messages and threads | |
| Tickets | Linear, Asana, Jira | Search issues and project tasks |
| Company Documents | Notion, Confluence, Google Drive | Search company documents and wikis |
| MCP Server | (custom) | Connect a real MCP server by URL |
MCP servers
Admins can also connect real Model Context Protocol servers. When adding an MCP Server connector, the admin provides a server URL, probes it to discover available tools, and adds it to the assessment. Candidates interact with MCP connectors the same way — the tools appear in their AI assistant automatically. MCP servers support optional authentication via static bearer tokens or OAuth flows.How candidates use connectors
During the setup phase, candidates browse available data sources and choose which to connect. During the task phase, they can search connected data sources through the AI assistant using natural language queries. The search is powered by TF-IDF ranking, returning relevant documents that the candidate can then reference in their deliverable.Signal and noise
Each data source is tagged as either signal (relevant to the task) or noise (a distractor). This distinction is invisible to the candidate — they see all data sources equally and must determine which ones are useful. The scoring engine tracks:- Which data sources the candidate connected
- Which documents they searched for and reviewed
- Whether they used signal documents in their deliverable
- Whether they included noise data uncritically
Candidate permissions
Admins can control what candidates can do with connectors via the Candidate Permissions card in the Data tab:| Permission | Default | Effect |
|---|---|---|
| Browse connector documents | On | Candidates can see individual documents within each connector |
| Add custom MCP servers | Off | Candidates can connect their own MCP servers |