ClickHouse's observability MCP server — investigate logs, traces and metrics with semantic tools rather than raw SQL, hosted or self-hosted.
8 views
Summary
Checkly's hosted MCP server gives an agent read access to your synthetic monitoring account — check health, failing results, test sessions, Rocky AI root-cause analyses, incidents and status pages — over streamable HTTP with OAuth.
Features
- Checkly-hosted streamable-HTTP endpoint at https://api.checklyhq.com/mcp — nothing to install or run
- OAuth for approved clients, or a Bearer API key with X-Checkly-Account for service identities
- Reads check state, check results, test sessions, result assets, incidents and status pages
- Starts and reads Rocky AI root-cause analyses on a failing check
- Triggers an existing check on demand to confirm a fix
- Granular scopes so an agent can be given read-only access to monitoring data
Installation
Set up this MCP server in your favorite AI agent — copy a ready-made configuration below.
Description
Checkly runs Playwright-based synthetic checks against production and tells you when a user journey breaks. The Checkly MCP Server is the first-party way to bring that signal into an agent session: instead of pasting a failing check's output into a chat, the agent queries the account directly while it is looking at the code.
A hosted server, not a package
There is nothing to install. The server is Checkly-operated at https://api.checklyhq.com/mcp, speaks streamable HTTP, and is reached the same way from any client that supports the transport:
claude mcp add --transport http checkly https://api.checklyhq.com/mcpOAuth is the recommended authentication path and covers Checkly-approved clients. Where OAuth is not an option — or where a workflow needs a stable service identity — a user or service API key works through an Authorization: Bearer <key> header, with X-Checkly-Account selecting which account to read when a key spans more than one.
What the agent can see
The tool surface follows the product: account membership, plan and feature entitlements; check state (passing, failing, muted, deactivated); check results and test-session data; result assets and normalised manifests; status pages and incidents; and account environment variables. It can start or read a Rocky AI root-cause analysis, and it can trigger an existing check on demand — useful for confirming that a fix actually cleared the failure rather than guessing.
Where it earns its place
The practical pattern is incident work. A check goes red, and the same session that holds your diff can pull the failing result, read the RCA, and re-run the check after the fix — no tab switching, no copy-pasting logs, and no separate deploy to observe. For teams already running Checkly's monitoring-as-code workflow, it closes the loop between the agent that wrote the change and the monitors that judge it.
Worth knowing before you connect
The scopes are granular and listed explicitly in Checkly's setup docs — reads for checks, results, RCA, test sessions, assets, incidents, status pages and usage, plus writes for checks, incidents, environment variables and account invites. Grant the narrow set your workflow needs; a monitoring account holds environment variables and secrets alongside the check data.
Related MCP Servers
Vanta's official remote MCP server — query controls, tests, vendors, vulnerabilities and policies across SOC 2 and ISO 27001, and remediate failing tests from your agent.
Turso's hosted MCP server — manage Turso Cloud databases, branch them point-in-time and run SQL from a coding agent, authorised over OAuth with no API token to paste.
Google's official Firebase MCP server — 70+ tools across Firestore, Auth, Crashlytics, Cloud Functions, Storage, Messaging and Remote Config, shipped inside the Firebase CLI.