MCP server
Aient is an MCP-native AI SRE
Aient exposes your production problems, OpenTelemetry traces, and its AI remediation loop over the Model Context Protocol. Connect from Claude, Cursor, or your own agent, ask what's broken in production — and open a reviewed GitHub fix PR, without leaving the chat.
Point your client at the endpoint
Aient's MCP server is remote and HTTP-based — nothing to install or self-host. Add the endpoint and your client walks you through sign-in. Claude Code:
claude mcp add --transport http aient https://aient.ai/mcpCursor needs a static auth block so Settings shows Authenticate. Paste this into .cursor/mcp.json:
{
"mcpServers": {
"aient": {
"url": "https://aient.ai/mcp",
"auth": {
"CLIENT_ID": "aient-mcp-cursor",
"scopes": [
"aient.mcp.read",
"aient.mcp.onboarding",
"aient.mcp.write"
]
}
}
}
}Codex uses dynamic registration: add the same URL, then codex mcp login aient. Do not paste Cursor's auth block into Codex.
Everything from triage to fix
The same capabilities as the Aient app, as tools an agent can call. Read tools need the read scope; state-changing tools need an explicit mutation scope — remediation and most writes require the write scope.
Problems
Triage the issues Aient has detected and grouped from your telemetry.
list_problemsget_problemupdate_problem_priorityacknowledge_problemmute_problemresolve_problem
Telemetry
Query the underlying OpenTelemetry data behind any problem.
query_tracesquery_logsget_spansdescribe_telemetrydetect_anomaliesservice_health_summary
Remediation
Drive the fix loop — each remediation opens a reviewed PR you merge.
request_problem_remediationretry_problem_remediation_attemptlist_active_remediationsget_problem_remediation_attempt
Setup & keys
Verify the connection, manage environments, and mint scoped keys.
verify_connectionget_activation_statuslist_environmentscreate_environment_keylist_aws_firehose_drainscreate_aws_firehose_drainlist_api_keyscreate_api_key
Plus environment keys, threads, onboarding, and more — discoverable by any client at /.well-known/mcp/server.json.
The agent proposes; you dispose
Access is read-only by default. Listing problems, querying traces, and reading service health all use the aient.mcp.read scope. An agent can understand your production without being able to change anything.
Anything that acts — request_problem_remediation above all — requires the explicit aient.mcp.write scope. And even then, remediation opens a reviewed GitHub pull request. It never merges for you. The merge button stays yours, exactly as in the web app.
Every organisation runs isolated, and the server advertises its OAuth protected-resource metadata so clients authenticate against the right boundary. See the pricing page for how base-plus-usage billing and budget ceilings apply to agent-triggered work.
Questions, answered
What is the Aient MCP server?
It is the same Aient AI SRE, reachable over the Model Context Protocol. Aient exposes your production problems, OpenTelemetry traces and logs, and its AI remediation loop as tools an agent can call — so an MCP client like Claude, Cursor, or your own agent can ask what's broken in production and act on it, without you opening the dashboard.
What can an agent actually do with it?
Read-only, it can list and inspect detected problems, query traces, logs and spans, detect anomalies, and summarise service health. With the write scope it can request a remediation, which spawns an agent that root-causes the issue against your source code and opens a reviewed GitHub pull request — you still review and merge. There are more than 30 tools in total.
How does an agent authenticate?
With scoped credentials. Read access uses the aient.mcp.read scope; mutations such as requesting a remediation require aient.mcp.write. The server advertises OAuth protected-resource metadata at /.well-known/oauth-protected-resource, so a compliant MCP client can complete the sign-in flow on its own.
Which MCP clients work?
Any client that supports a remote (HTTP) MCP server: Claude Desktop and Claude Code, Cursor, and your own agents. You point the client at https://aient.ai/mcp and it prompts you to sign in. There is nothing to install or self-host.
Is it safe to give an agent access to production?
It is read-only unless you explicitly grant the write scope, the remediation tool opens a pull request that merges only after explicit human approval on the exact PR head (merge automation is off by default), and each organisation's data is isolated. The agent proposes; you dispose — nothing merges without you.
Does using the MCP server cost extra?
No separate fee — it's part of Aient ($99/month base plus usage). Reading problems and telemetry over MCP is included; the remediation work an agent triggers is billed as usage against the budget you set, with live spend tracking.
Put an AI SRE in your agent's toolbox
Connect Aient over MCP and let your agents see — and fix — what's breaking in production. You review every pull request before it ships.