WorkSights · Developer documentation
The WorkSights MCP Server
Connect Claude — or any client that speaks the Model Context Protocol — directly to WorkSights. Your assistant gets the same view of your data you have in the product: nothing more, nothing less.
https://app.worksights.ai/mcpQuick start
Claude (web and desktop)
- Open Settings → Connectors → Add custom connector.
- Paste
https://app.worksights.ai/mcpand continue. - Sign in with your usual WorkSights single sign-on and approve the connection. That's it — WorkSights's tools appear in your conversations.
Claude Code
claude mcp add --transport http worksights https://app.worksights.ai/mcpOther MCP clients
The server speaks stateless Streamable HTTP with OAuth 2.1 authorization (PKCE and dynamic client registration). Clients discover the authorization server automatically via /.well-known/oauth-protected-resource/mcp — no API keys, no manual configuration.
What you can ask
- “What has my team been working on over the last two weeks?”
- “Summarize what Alex worked on this month, day by day.”
- “How does this sprint compare to the last one for the engineering group?”
- “Ask WorkSights: who on my team is closest to burning out?”
Under the hood, four kinds of object come back: activity (row-level data), score (daily numbers with a narrative summary), memory (pre-computed synthesis at daily, weekly, and monthly granularity), and observation (observer outputs that distill into memory). The get_insights tool chooses the right granularity for the question so long ranges stay digestible — your assistant does the storytelling from there.
Tools
| Tool | What it does |
|---|---|
| whoami | Who you are: your profile, the accounts you can access, and the limits on this connection. The starting point when an assistant needs to pick an account. |
| search_users | Find people on your team by name or email — limited to the users you are allowed to see. |
| get_user | One user in full: profile, soul, status, product access, and their active categories. |
| get_insights | The primary tool. Give it people and a timeframe; it picks the right level of pre-computed memory (daily, weekly, or monthly) to fit the range and returns them stitched chronologically. Omit the people to cover everyone you can see. |
| get_activities | Raw activity rows for one or more users and a timeframe — the row-level record of what happened. |
| get_scores | Daily scores, including the narrative summary where the date allows. Omit users for your whole visible team. |
| get_memories | Raw memory rows at an explicit granularity — the pre-computed synthesis get_insights chooses from. |
| get_observations | Observer outputs about your team for a period. |
| ask_worksights | Ask WorkSights itself a free-form question through its chat engine and get the completed answer. Supports my-team and all-users subjects, and multi-turn conversations. |
Access, security, and limits
- Your permissions, exactly. Every call runs as you: the connection sees only the people, clients, and data your WorkSights account can see. Connections are read-only.
- Multiple accounts. Calls default to your default account; every tool accepts an
account_id, andwhoamilists the accounts you can choose from. - Rate limits. 60 requests per minute per connection, and at most 10 concurrent
ask_*questions. - Audit trail. Every request — including each individual tool call — is logged with your identity and the connected application.
- Revoking access. Disconnect from your AI assistant's connector settings at any time; tokens are no longer honored once revoked.