Checkmk now ships its own MCP server. Since 2.5.0p13 (September 2026), a site in the Pro, Ultimate, Ultimate MT or Cloud edition can serve AI agents directly — read-only, marked experimental, and bound to the permissions of the Checkmk user who signs in. It is not part of the Community edition, and it does not write: no downtimes, no acknowledgements, no configuration changes. For those, and for Checkmk behind one endpoint together with other systems, the options are community projects, one commercial hosted service, or a gateway.
This page lists all six MCP options we could find for Checkmk, the built-in server included, and puts them side by side across ten dimensions. We make one of them, which is why the sourcing rules below matter: no scores, no stars, no rankings — every cell states a fact and links to the evidence.
The real choice is not the tool count but the shape of access: the vendor’s built-in read-only server, a small Checkmk-specific agent connector, a hosted endpoint with nothing to operate, or one governance layer across Checkmk and further systems. As of 23 October 2026 the field looks like this — six MCP options, plus three adjacent projects: two that are not MCP servers, and one announced server that can no longer be verified. Every claim in the table is a summary of sourced cells in the comparison matrix below.
Read-only status access with nothing to configure wrong — and the only dedicated server here that can verify callers' OAuth 2.0 JWTs against your own identity provider
No — read-only by design; eight tools are the entire surface
No writes at all; repository created and last pushed on the same day in March 2026
Broad writes against one Checkmk from a Python process your LLM client launches, or from one shared instance over HTTP
Yes — 90 of its 154 tools write or delete, from hosts and rules to users and passwords
Self-declared alpha; hosted over HTTP, one shared token grants every tool, with no per-user authorization
Docs review
ToolMesh + the Checkmk DADL is the only option here that publishes a measurement of its reach into the Checkmk REST API — 146 tools across 63% of the endpoints, including the Event Console, Business Intelligence and broad Setup coverage. vibeMK documents 154 tools across monitoring and Setup but no endpoint coverage, so the two cannot be ranked by reach. Checkmk’s built-in server reads monitoring state, history, Event Console events and the Setup configuration, but publishes no tool count and writes nothing. Among the plain local processes, the Checkmk MCP Server (jlk) and vibeMK both reach the Event Console; only the Checkmk MCP Server (jlk) also documents Business Intelligence.
A deliberately small, read-only MCP server: eight tools for hosts, services, downtimes, folders and tags. It runs as a central HTTP service rather than as a per-user subprocess, and among the dedicated Checkmk servers here it is the only one with optional OAuth 2.0 JWT verification of callers against an external identity provider. Repository created and last pushed on the same day in March 2026.
Checkmk's own MCP server, built into the site since 2.5.0p13 (September 2026) and included at no extra cost in Pro, Ultimate, Ultimate MT and Cloud — not in the Community edition. It is disabled by default and marked experimental. Clients connect over Streamable HTTP and sign in as a Checkmk user through the site's own OAuth 2.0 server, so the agent works with that user's permissions and no automation secret leaves the site. Strictly read-only: hosts, services, downtimes, comments, Event Console events, availability, metric history, and the Setup configuration with its change history, aimed at root-cause analysis. Checkmk does not publish a tool count.
37 tools in eight categories, and the broadest coverage of Checkmk's specialised subsystems in this list: Event Console, Metrics API and Business Intelligence are all addressed. Self-described as production ready; the last commit is from September 2025.
The only option hosted by a third party: 68 pre-built actions behind a StackOne endpoint, with OAuth token exchange, storage and refresh handled by the vendor. Feature-wise it is the closest thing here to a gateway — it is the only other option with dynamic tool discovery, a code mode, and a classifier for prompt injection in tool responses. No installation, but your Checkmk credentials and monitoring data pass through StackOne's infrastructure.
Not a Checkmk-specific server but a self-hosted gateway that reads a declarative YAML description of the Checkmk REST API — 146 tools plus three composites, 63% of the API surface, including the Event Console and Business Intelligence since 14 August 2026. Governance (per-tool authorization, server-side credentials, output policies, audit) comes from the gateway rather than from the connector. Still uncovered: HW/SW inventory, agent registration, DCD, LDAP connections, and historical Event Console events, which Checkmk only exposes on an unstable API version.
The most visible community project — announced in the Checkmk forum, where Checkmk staff responded positively. Among the dedicated Checkmk servers here it documents write access across the most configuration objects: hosts, folders, rules, users, groups, passwords, tags and time periods. Its install guide counts 154 tools, 64 of them read-only. Since October 2026 it is on PyPI, has a read-only mode, and can also run as one shared HTTP service behind a bearer token. The README still labels it alpha.
Rows are the ten dimensions; columns are the options. The table scrolls horizontally. Superscript numbers link to the source for that cell.
“Not documented” means the sources we checked do not state it — not that the feature is absent. If you maintain one of these projects and we got something wrong, open an issue and we will fix the data file.
Verified on a community (Raw) edition site, including the Event Console and BI, which Checkmk registers for every edition; CEE-only endpoints such as agent bakery and signature keys are not covered5
Raw fully supported, BI and Event Console endpoints included; Agent Bakery and custom graphs need Enterprise or Cloud6
Tagged releases and container images; the Checkmk description is a versioned YAML file in a public registry5
Official PyPI package vibemk (versions pinnable) and an entry in the official MCP Registry, or git clone; an older third-party copy on PyPI, iflow-mcp_chexma-vibemk, carries the August 2025 code and is not the maintainer's channel18
68 actions, but tool discovery and a "Code Mode" that keeps raw responses in a sandbox hold the loaded context down — the vendor reports massive token savings21
146 tools, but Code Mode replaces the catalog with two meta-tools, keeping the loaded context near-flat — massive token savings, with the exact rate depending on the workload22
Explicit — tools for listing and activating pending changes plus a composite that applies them; the Monitoring/Setup split is documented for the model5
Explicit — tools to list and activate pending changes; NEVER_ACTIVATE_CHANGES=true blocks activation11
A gateway that speaks the Checkmk REST API can only ever reach what the REST API exposes. Several things in Checkmk sit outside it, and for those a purpose-built server — or a Checkmk plugin — will always win:
Livestatus. The monitoring core answers on a socket, not over REST. For large or frequent status queries, Livestatus is dramatically cheaper and exposes columns and filters the REST layer does not surface. A dedicated server can open that socket directly; a REST-based option cannot.
The Event Console.mkeventd has its own protocol, and the REST surface for events is only partial: current events can be listed, acknowledged and archived, but historical events are exposed solely on an API version Checkmk marks unstable, which none of the third-party options here reaches. Four options read current events, Checkmk’s built-in server among them; whether the built-in server also reaches the history is not documented.
Business Intelligence. BI aggregations are compiled server-side. Reading them meaningfully means understanding the aggregation tree, not just fetching JSON.
The site filesystem and the cmk CLI. Check plugin development, local checks, agent baking internals, omd operations and configuration files are reachable over SSH or on the site, never over REST.
Notifications and events as a stream. Checkmk pushes notifications through notification handlers. There is no REST subscription to poll, so anything event-driven needs to run as a handler inside Checkmk.
Performance data over long ranges. Pulling months of metrics through the REST API is far more expensive than reading the RRDs where they live.
If your use case is any of the above, a REST gateway is the wrong shape regardless of who makes it.
A gateway earns its keep when Checkmk is not the only system in the picture:
Questions that span systems. “Which hosts are critical in Checkmk, and who owns them in NetBox?” is one agent turn against one endpoint, instead of two connectors that cannot see each other. The public DADL registry lists 30 APIs and over 4,000 tools as of August 2026, so the second system is often already described.
One place for credentials. The automation secret lives server-side and is injected at call time, rather than sitting in every operator’s client configuration.
One authorization model and one audit trail across every connected system, instead of per-connector conventions.
Context economy at scale. 146 Checkmk tools is a lot of schema for a model to carry; a catalog that stays flat matters more the more systems you connect. This one is not ours alone — StackOne solves the same problem for its hosted catalog.
The honest trade-off: a gateway is an additional component you have to run, monitor and keep available — and it only pays for itself once more than one system is behind it. If Checkmk is the only thing you will ever connect, and read-only queries are all you need, the leaner answer on a commercial edition is Checkmk’s built-in server — nothing to install and nothing extra to run. On the Community edition, CheckMCP exists and is eight tools.
The machine-readable coverage report for our option — every tool, its access level, and which parts of the Checkmk REST API are and are not covered — is at dadl.ai/d/checkmk. It records 146 tools plus three composites, 63% of roughly 230 endpoints, with the uncovered areas named explicitly.
Yes, since September 2026. Checkmk 2.5.0p13 added a built-in MCP server to the Pro, Ultimate, Ultimate MT and Cloud editions at no extra cost; the Community edition does not include it. It is disabled by default, marked experimental, and strictly read-only: an agent can query hosts, services, downtimes, comments, Event Console events, availability, metric history and the Setup configuration, but cannot schedule downtimes, acknowledge problems or change the configuration. Clients connect over Streamable HTTP and sign in through the site's own OAuth 2.0 server, so the agent only sees what the signed-in user may see. Checkmk's earlier 'Explain with AI' feature (Checkmk 2.5, April 2026) is a different thing: it runs inside the GUI of the Cloud edition and cannot be called by an external agent.
Do I still need a third-party option now that Checkmk ships its own server?
Only for what the built-in server does not do. Writes need one of the other options: the Checkmk MCP Server (jlk) documents downtimes and acknowledgements, and vibeMK, StackOne and ToolMesh also document configuration changes. On the Community edition the built-in server is not available at all. And if Checkmk is one of several systems an agent should reach under one authorization model and one audit trail, that is a gateway's job. For read-only analysis on a commercial edition, the built-in server is the natural first choice.
Can an AI agent write to Checkmk through MCP, or only read?
Both, depending on the option. CheckMCP and Checkmk's built-in server are read-only by design. The other four support writes at different depths: the Checkmk MCP Server (jlk) documents downtimes and acknowledgements, while vibeMK, StackOne and ToolMesh also document configuration writes such as creating hosts and folders or editing rules. Note that Checkmk splits the two worlds: monitoring actions such as downtimes take effect immediately, while Setup changes are staged as pending changes and do nothing until they are activated.
Which option covers the Event Console and Business Intelligence?
Two of them. The Checkmk MCP Server (jlk/checkmk-llm-server) documents Event Console, Metrics API and Business Intelligence coverage. The ToolMesh Checkmk DADL added both on 14 August 2026 — seven Event Console and fourteen BI tools, verified against a live Checkmk 2.5.0p11 site. vibeMK covers the Event Console but has no BI tool; since October 2026 its README places both in every edition, Raw included. Checkmk's built-in server reads Event Console events, but BI is not among its documented areas. CheckMCP covers neither. One limit applies to every third-party option: historical Event Console events are only exposed on an API version Checkmk marks unstable, so none of them reaches them; for the built-in server this is not documented.
Where do my Checkmk credentials end up?
That is one of the clearest differences between the options. Checkmk's built-in server needs no automation secret at all: each user signs in through the site's own OAuth server, and the client holds a token bound to that user's permissions that expires after 48 hours. The Checkmk MCP Server (jlk) reads the automation user's secret from the LLM client's configuration or a .env file on the operator's machine. vibeMK reads it from its environment: the client configuration when the client launches it, the server's environment when it is hosted centrally over HTTP. CheckMCP reads it from a YAML file on the host it runs on. StackOne stores and refreshes the credential in its own cloud, and your monitoring data traverses StackOne's infrastructure. ToolMesh keeps it server-side in the gateway and injects it at call time, so it never appears in a client configuration.
Do these work with a self-hosted Checkmk behind a firewall?
The five self-hosted options do, because they run inside your network; Checkmk's built-in server runs inside the site itself, but its OAuth sign-in requires the site to be reachable over HTTPS. For the hosted StackOne connector this is not documented, and a vendor-operated endpoint needs some way to reach your Checkmk site.
Which Checkmk versions are supported?
It varies and matters. Checkmk's built-in server first shipped in 2.5.0p13, and Checkmk's setup guide asks for 2.5.0p14 or newer in a commercial edition. CheckMCP is tested against Raw Edition 2.3.0p23. The Checkmk MCP Server (jlk) requires 2.4.0 or newer. vibeMK reports full support for 2.5.x, 2.4.x and 2.3.x, tested against 2.5.0p14, with 2.2 and older unsupported. The ToolMesh DADL targets 2.4 and 2.5 and was live-verified on 2.5.0p11. StackOne does not document version support.
Reference from here on — the adjacent projects, the methodology and disclosure, the changelog and the sources. Nothing below is needed for the decision above; it is where the evidence lives.
Built into Checkmk 2.5 (April 2026) and limited to the Cloud edition. It explains alerts inside the Checkmk GUI; it is not an MCP server and cannot be called by an external agent.
A commercial AI assistant plugin for the Checkmk GUI by Lynxmind, using a bring-your-own-LLM model. It talks to LLM APIs directly and does not speak MCP.
Announced in a 2025 blog post with a pip install command, but the linked GitHub repository returns HTTP 404 as of 13 August 2026, so nothing could be verified. Listed here so the absence is documented rather than silent.
Disclosure. This page is published by Dunkel Cloud GmbH, which makes ToolMesh — one of the six options listed. We have an obvious interest in how it looks. The countermeasures are structural: options are ordered alphabetically, every cell carries a source link, no cell contains a score or a ranking, and the two “when to choose” sections above name the cases where our option is the wrong tool.
How the facts were gathered. Each option’s own repository, README and documentation, read on 13 August 2026. Repository metadata — licence, last commit, stars and forks — comes from the GitHub API on the same date. vibeMK was read again on 21 October 2026, after its maintainer reported changes; its row and its metadata reflect that date. Checkmk’s built-in server was added on 23 October 2026 from Checkmk’s release notes (Werks #22263 and #22264), its launch post and the setup guide in its knowledge base; as part of the commercial editions it has no public repository metadata. Where a project does not state something, the cell says “not documented” rather than guessing.
For hosted products that means the platform documentation, not only the per-connector page. A connector page describes the connector; capabilities that span the whole product — tool discovery, how responses are handled, screening for injected content — are documented elsewhere. Reading only the connector page understates such products, and it understates them in exactly the areas where a gateway would claim strength.
Depth of verification. The five options we do not maintain are documentation review, not test results — read, not run. Vendor figures we cannot count ourselves are quoted as such. Our own entry is the exception: the Checkmk DADL was live-verified against a Checkmk 2.5.0p11 site on 13 August 2026, including a full write cycle, and re-verified on 14 August after Event Console and BI coverage were added. The maintainers of the listed projects were notified when this page went live; corrections land in the data file.
Corrections and additions. The matrix lives as one YAML file in this site’s repository; the site repository itself is private, so corrections travel as issues on the public ToolMesh repository. Correcting a fact, or adding an option we missed, takes one issue. Maintainers: if we have misrepresented your project, that is a bug and we will fix it.
Checkmk's built-in MCP server added as a sixth option. Since 2.5.0p13 (September 2026) Checkmk ships a read-only, experimental MCP server in its commercial editions; this page had still stated that no official server existed. Introduction, decision table, FAQ and methodology were updated accordingly, with a new question on when a third-party option is still needed. The community project jlk/checkmk-llm-server is now listed as Checkmk MCP Server (jlk) to keep it apart from Checkmk's own server, and the claims that CheckMCP and StackOne were the only ones of their kind were narrowed where the new option also qualifies.
vibeMK updated after its maintainer reported changes. Releases 0.5 to 0.6.4 (October 2026) added a PyPI package, a read-only mode and an optional HTTP transport, and the README's edition table now places BI and the Event Console in Raw. Its row was re-read against the repository on 21 October 2026, tool count (154), versions, transport, credentials and distribution included. Statements that relied on vibeMK running only as a local process were corrected, among them CheckMCP being the only central HTTP service, and our own coverage claim now rests on the published measurement alone.
First publication. Five MCP options compared across the ten dimensions, plus three adjacent projects: two that are not MCP servers and one announced server that could not be verified. Facts taken from each project's own repository and documentation; our own entry additionally verified against a live Checkmk 2.5.0p11 site.