Connections

A connection is one system your agents may reach: an MCP server, the credentials it is called with, and a mode for each tool it offers.

Adding one

  1. 1

    Pick it from the catalog, or paste a URL

    The catalog carries the endpoint and auth style for the systems we have checked. Anything else works too. Paste the MCP server URL and Tier Two probes it for its transport, its auth requirement, and its tool list.

  2. 2

    Choose how it authenticates

    Either each person signs in with their own account, or the workspace provisions one shared key. This is the single most consequential choice on the page. See below.

  3. 3

    Tier Two pins the tools

    The tool list is read once and stored, so a server that adds a tool tomorrow does not silently widen what your agents can do. Refresh tools when you want the new ones, and review the diff.

  4. 4

    Apply a posture

    Recommended is pre-selected: reads run, writes wait for a person. You can change any single tool afterwards.

Who the call runs as

AuthThe call runs asUse it when
Sign in with your accountthe person using the agenteveryone has their own seat downstream, and you want that system's own permissions to keep applying
Shared keyone workspace credential, forwarded server-sideseats are scarce or setup per person is not worth it. Tier Two's modes become the only limit, so keep them tight
Personal API keythe person's own key for that systemthe system has no OAuth but people do have individual keys
A shared workspace key is decrypted only for the call that needs it, used in process to forward that call, and never sent to a device. But it is still one credential for everyone: what stops a person from doing more with it is the mode you set on each tool.

What we have checked

The catalog records how each entry was verified (a live probe, the vendor’s own documentation, or a secondary source) and the date. Entries we have not been able to verify say so in the product rather than quietly failing later.

SystemAuthVerified
PostHogSign in with your accountdocs · 2026-09-07
LinearSign in with your accountdocs · 2026-09-02
SentrySign in with your accountdocs · 2026-09-02
GitHubSign in with your accountdocs · 2026-09-02
NotionSign in with your accountdocs · 2026-09-02
AtlassianSign in with your accountdocs · 2026-09-02
StripeSign in with your accountdocs · 2026-09-02
CloudflareSign in with your accountdocs · 2026-09-02
SupabaseSign in with your accountdocs · 2026-09-02
NeonSign in with your accountdocs · 2026-09-02
FigmaSign in with your accountdocs · 2026-09-02
MetabaseShared keyprobe · 2026-09-02
Hugging FacePersonal API keydocs · 2026-09-02
Context7Personal API keydocs · 2026-09-02
VercelSign in with your accountnot verified yet
HubSpotSign in with your accountnot verified yet

Naming and slugs

  • A connection’s slug is what the agent sees: linear__create_issue comes from the slug linear.
  • Renaming the display name is safe; changing a slug changes every tool name with it.
  • Disabling a connection hides its tools from every agent immediately, without deleting its history.