Connect Cursor to a team knowledge base
Point Cursor at your team's base over MCP and it stops coding from memory alone: it searches your snippets, saved API calls and runbooks while it works, reuses the versions your team already approved, and files new ones back for review. The connection is a workspace API key and one JSON block in .cursor/mcp.json — no plugin, no sync daemon. The payoff compounds with team size: every developer's Cursor querying one reviewed base means the canonical retry helper, the blessed deploy runbook and the correct auth curl are what agents reach for, on every machine, instead of five private approximations.
How it works
- Generate a workspace API key in Settings → API keys — keys are free on every plan; agent writes (create and update) need Pro & Team, $8/user/month.
- Add The-Snip to .cursor/mcp.json with the config below; Cursor lists the eight tools once it reconnects.
- Add a project rule so Cursor uses it consistently: search the-snip before writing utility code or calling internal APIs; propose new reusable snippets with create_snippet.
- Code as usual — Cursor calls search_base and pulls your canonical snippets and saved requests into context as it works.
- Review what it proposes: approve, edit or reject from the review queue. Approved items are instantly searchable by every teammate and agent.
{
"mcpServers": {
"the-snip": {
"url": "https://the-snip.com/mcp",
"headers": { "Authorization": "Bearer snip_········_····" }
}
}
}Questions, answered.
Do agents need their own seats?
No. Agents authenticate with workspace API keys, not seats — you pay per human on the Pro & Team plan. Issue one key per developer or per agent so keys can be revoked individually.
What does Cursor actually get back from search_base?
Ranked matches from your workspace's approved canon — snippets with their code and language, saved API calls, and docs — weighted so title matches beat body mentions. It's the same search your team uses in the app, over MCP.
Can Cursor write to the base without a human in the loop?
It can propose — create_snippet, create_api, create_doc, update_item — but nothing becomes canon until a reviewer approves it. That review gate is the product's core promise: agents contribute, humans decide.
We use Claude Code too — same setup?
Same server, same key, same eight tools; only the config file differs (.mcp.json for Claude Code, .cursor/mcp.json for Cursor). One workspace serves every MCP client your team runs.
What happens if a key leaks?
Revoke it in Settings → API keys and issue a new one — keys are workspace-scoped and hashed at rest, so one key never exposes another workspace. Scoped, revocable keys are why agents don't share your human login.
Does search cover docs and saved API calls, or just snippets?
All three item types — snippets, saved API calls and Markdown docs — come back from one search_base call, ranked together. Cursor can pull the runbook and the request shape in the same query it uses to find the helper.
Start your base — free.
Free: 25 items, no card. Pro & Team: $8/user/mo — unlimited items, REST API, review workflow, and the hosted MCP server.