Stable release 1.2.2. Hosted MCP and CLI browser login are enabled in Production DE and US. See release status and ChatGPT setup.
The server delegates business operations to the official API v1 client and exposes 208 tools: 84 reads and 124 writes. It supports local stdio and a regional HTTP runtime with OAuth. Each request checks current permissions.
Choose your connection
| Host or workflow | Start here |
|---|---|
| ChatGPT with hosted OAuth | Connect TeamGrid to ChatGPT; no local installation |
| Local Codex or another stdio host | Configure a local host and first read |
| Make and verify a change | Safe task-write walkthrough |
| Files, exports, large documents or protocol integration | Resources and protocol support |
Hosted Production advertises full with writes enabled. The local default is
core, which advertises only reads. Visibility, granted scopes and permission to
act are separate: approving a connection does not give a person new TeamGrid rights.
Choose a workflow
| Need | Profile or entry point |
|---|---|
| Basic inspection | core: 22 operational reads; the default |
| Contacts and colleagues | collaboration: 29 reads |
| Configuration inspection | governance: 28 reads |
| Existing broad read surface | The all profile exposes 36 curated reads |
| Personal work context | context |
| Task, time and comment work | work |
| A specific business domain | One of the eleven domain profiles in the reference |
| All reviewed business tools | full: 208 tools, with current scopes and write gates still enforced |
all keeps its existing meaning. Profiles and allow/deny filters select tools;
they do not add permissions or bypass sharing, workspace locks or revision checks.
Use the smallest profile suitable for the workflow.
Start locally
Install the matching stable packages:
npm install --global @teamgrid/cli@1.2.2 @teamgrid/mcp-server@1.2.2
teamgrid auth loginSign in, select one workspace and approve the requested scopes; see browser login. The stdio process uses the selected CLI credential store and opens no browser itself.
Results and boundaries
Reads are bounded snapshots. Follow opaque cursors for complete listings; a search result is not a full inventory. Workspace context identifies the acting person when applicable and the effective time zone. Service credentials do not represent an invented human user.
Writes require the exact workspace and any declared revision or idempotency key. Receipts distinguish completed, accepted, partial and uncertain outcomes. Inspect an uncertain result before retrying. Multi-tool workflows are not one transaction.
Large documents use revision-bound sections. Private files and exports are available through bounded, freshly authorized resource reads; transfer URLs remain internal. Credential secrets, arbitrary HTTP or database access and durable synchronization streams stay outside the model tool surface.
Run a first read, configure a host, browse the 208-tool reference or review security and writes.