Connect an AI agent (MCP)
Uptizm runs a remote MCP server at https://mcp.uptizm.com, so an AI agent can read and manage your team in plain language: list monitors, look at checks and incidents, add custom metrics, build your board. It works with any MCP client that speaks Streamable HTTP. A client that supports MCP sign-in opens Uptizm in the browser and you approve it there; any other client sends a team API key.
Sign in with your Uptizm account #
Add the server to your client, then start its sign-in. The browser opens Uptizm. Sign in as usual (two-factor and Google, GitHub or Apple sign-in work as they do in the app) and the consent screen shows which app is asking and where it will send you back, then asks two things:
- Team: the connection acts on the team shown. Switch team on that screen before you approve to connect another one.
- Access: read only, read and write, or full. These are the presets of Authentication, with the same abilities.
Approve, and the browser hands control back to the client.
| Client | Add the server | Sign in |
|---|---|---|
| Claude Code | claude mcp add --transport http uptizm https://mcp.uptizm.com |
/mcp, then pick uptizm |
| Claude Code, with the Uptizm skill | /plugin marketplace add uptizm/skills, then /plugin install uptizm@uptizm |
/mcp, then pick plugin:uptizm:uptizm |
| claude.ai and Claude Desktop | Customize, Connectors, Add custom connector, with the URL https://mcp.uptizm.com |
Connect on the connector |
| ChatGPT | With developer mode on, Plugins, Add custom MCP server, with the URL https://mcp.uptizm.com and OAuth |
Connect on the server |
| Codex | In ~/.codex/config.toml: [mcp_servers.uptizm] with url = "https://mcp.uptizm.com" |
codex mcp login uptizm |
| OpenCode | In opencode.json: "mcp": {"uptizm": {"type": "remote", "url": "https://mcp.uptizm.com"}} |
opencode mcp auth uptizm |
| Cursor | In ~/.cursor/mcp.json: "mcpServers": {"uptizm": {"url": "https://mcp.uptizm.com"}} |
Connect beside the server in the MCP settings |
| VS Code | In mcp.json: "servers": {"uptizm": {"type": "http", "url": "https://mcp.uptizm.com"}} |
Start above the server, then allow the sign-in |
| Gemini CLI | In ~/.gemini/settings.json: "mcpServers": {"uptizm": {"httpUrl": "https://mcp.uptizm.com"}} |
/mcp auth uptizm |
| Zed | In settings.json: "context_servers": {"uptizm": {"url": "https://mcp.uptizm.com"}} |
Zed asks when the server starts |
Use an API key instead #
A client without MCP sign-in, or a script, sends a team API key as a bearer token. Create a key in Settings, Team, API keys with the access the agent needs, keep it in an environment variable such as UPTIZM_API_KEY, and add the header to the server's entry:
Authorization: Bearer upz_...
Claude Code, which stores the header with the key in its own config:
claude mcp add --transport http uptizm https://mcp.uptizm.com --header "Authorization: Bearer ${UPTIZM_API_KEY}"
Codex, in ~/.codex/config.toml:
[mcp_servers.uptizm]
url = "https://mcp.uptizm.com"
bearer_token_env_var = "UPTIZM_API_KEY"
Cursor, in ~/.cursor/mcp.json:
{
"mcpServers": {
"uptizm": {
"url": "https://mcp.uptizm.com",
"headers": { "Authorization": "Bearer ${env:UPTIZM_API_KEY}" }
}
}
}
VS Code, in mcp.json, asking for the key once and storing it securely:
{
"inputs": [
{ "type": "promptString", "id": "uptizm-key", "description": "Uptizm API key", "password": true }
],
"servers": {
"uptizm": {
"type": "http",
"url": "https://mcp.uptizm.com",
"headers": { "Authorization": "Bearer ${input:uptizm-key}" }
}
}
}
OpenCode, Gemini CLI, Windsurf and Zed take the same headers object on the server's entry (Windsurf names the URL serverUrl). ChatGPT signs in with OAuth only.
How a connection acts #
Approving the consent screen creates a team API key named after the client, such as "Claude Code (MCP)", with the access you picked. Every tool call runs as that key, or as the key you sent yourself, so everything Authentication, Errors and Rate limits say about keys holds here too: a tool the access level does not cover answers with the missing ability, a row of another team is not found, and the key's per-minute budget applies.
- The key counts toward your plan's key limit. Signing in again from a client that kept its registration (Claude Code, Codex and the other local clients) replaces its key instead of adding one. A client that registers anew on each connection, as claude.ai does, gets a new key; the old one is revoked once its sign-in expires, or revoke it yourself.
- To disconnect, revoke the key in Settings, Team, API keys. A signed-in client then asks you to sign in again.
- Board tools act on the board of the person who approved the connection or created the key.
- A signed-in connection unused for 30 days expires, and its key is revoked.
Tools #
Each tool runs one API endpoint; the API reference documents the fields. Tools that only read are marked read only, so the client can ask before a change.
| Area | Tools |
|---|---|
| Account | AccountGet, MemberList |
| Dashboard | DashboardStats, DashboardIncidents, DashboardMonitors, SuggestionList |
| Monitors | MonitorList, MonitorGet, MonitorCreate, MonitorUpdate, MonitorDelete, MonitorPause, MonitorResume, MonitorCheckNow, MonitorAnalyze, MonitorAnalyzeResult |
| Checks | CheckList, CheckGet, CheckEpisodeList, MonitorUptime, MonitorResponseTimes, MonitorAnomalies, MonitorStatusHistory, MonitorMaintenanceSpans, ContentVersionList, MetricCandidates |
| Custom metrics | MetricList, MetricCreate, MetricUpdate, MetricDelete, MetricReorder, MetricPreview, MetricDiscover, MetricSeries, MetricHistory, MetricReadings |
| Incidents | IncidentList, IncidentGet, IncidentCreate, IncidentAcknowledge, IncidentResolve, IncidentReopen, IncidentAssign, IncidentPostmortem, IncidentPostUpdate, IncidentAnalysis, IncidentAnalysisFeedback, IncidentDraftUpdate, IncidentDraftPostmortem, IncidentDigest, SuggestionAccept, SuggestionDismiss |
| Board | BoardGet, BoardReplace, BoardData |
| Status pages | StatusPageList, StatusPageGet, StatusPageCreate, StatusPageUpdate, StatusPageDelete, StatusPageMonitorAdd, StatusPageMonitorRemove, StatusPageMonitorReorder |
| Maintenance | MaintenanceList, MaintenanceGet, MaintenanceCreate, MaintenanceUpdate, MaintenanceDelete |
| On-call | OnCallNow, ScheduleList, ScheduleGet, ScheduleCreate, ScheduleUpdate, ScheduleDelete, RotationAdd, RotationRemove, RotationReorder, OverrideAdd, OverrideRemove, EscalationPolicyList, EscalationPolicyGet, EscalationPolicyCreate, EscalationPolicyUpdate, EscalationPolicyDelete, EscalationStepAdd, EscalationStepRemove, EscalationStepReorder |
| Notification channels | ChannelList, ChannelGet, ChannelCreate, ChannelUpdate, ChannelDelete |
[!NOTE]
MonitorAnalyze,MetricDiscover,IncidentAnalysisand the two draft tools spend your team's AI budget, as the same actions do in the app.IncidentCreate,IncidentResolveandMaintenanceCreatenotify people: channels, on-call or status page subscribers.
Managing members, invitations, status page subscribers, logos and API keys stays in the app.
The Uptizm agent skill #
The Uptizm skill teaches an agent how to use Uptizm well: which tool fits a task, what to confirm with you before a change, and what needs care on a live account. It uses this server when the agent is connected to it, and the REST API with UPTIZM_API_KEY otherwise. The Claude Code plugin above installs it with the server; any agent that reads agent skills installs it with:
npx skills add uptizm/skills
The skill's source is on GitHub.