# Fix list: ClickHouse MCP Server

From Anchor Terminal's listing at https://www.anchorterminal.com/tools/clickhouse-mcp-server, the October 2026 research run, assessed 9 October 2026. Grade B, 64.6 out of 100.

This is everything the published grade says the listing lacks, the biggest possible gain to the total first. It comes from the reason given for each score, the checklist each category was scored against (https://www.anchorterminal.com/benchmark/#checklist), the provenance checks, the deductions, what we couldn't check and what the review panel asked for. A fix counts at the next check, once it's public.

For a coding agent working on ClickHouse MCP Server: work through the items below in the product, its docs and its public pages. Each category gives the reason for its score, with the points each checklist item earned, and the checklist itself, so the gap is the items that earned less than their points. Change the product, not the wording, and keep a note of what you changed and where it's published.

## 1. Security & auth, 66 out of 100, up to 6 more on the total

Why it scored 66: The server holds one ClickHouse user and password, or an X.509 client certificate since 0.7.0, which ClickHouse grants and roles can scope and an admin can revoke. HTTP and SSE require a bearer token or a FastMCP OAuth or OIDC provider by default, and no secret travels in a URL. Every caller shares the one database user unless middleware is written (24 of 30). Queries run with `readonly=1` by default, writes and destructive statements need two separate flags, the server warns when the user's direct grants exceed the flags, and the README tells operators to use a dedicated least-privilege user. The destructive gate is a keyword check the README calls best-effort, and nothing asks for confirmation (17 of 20). Query results reach the model as stored. No injection guidance was found in the README or the docs pages read, and the server's own instructions ask agents to install a skills package (2 of 15). Each query's SQL is logged at INFO, ClickHouse's `system.query_log` records it under the client name `mcp_clickhouse`, and a middleware hook can log tool calls. No per-caller identity reaches ClickHouse (9 of 15). ClickHouse has a valid security.txt, a Bugcrowd bounty programme, and SOC 2 Type 2 and ISO 27001 on its trust centre. This repository has no SECURITY.md and no published advisories, though 0.3.0 and 0.5.0 carried security fixes (14 of 20).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-security):

- 0 to 30, the credential model. 30 for OAuth 2.1 with scopes, or scoped and revocable keys with rotation. 20 for plain revocable API keys. 10 for one all-powerful key. 10 off when a secret can travel in a URL query string as a documented option.
- 0 to 20, read-only or least-privilege modes, and confirmation or approval for destructive actions.
- 0 to 15, prompt-injection posture where the tool returns untrusted content (documented mitigations or guidance). A tool that returns no untrusted content gets 10.
- 0 to 15, audit logs or per-call visibility for the operator.
- 0 to 20, a security programme. security.txt or a disclosure policy, a bug bounty, SOC 2 or ISO 27001, advisories handled in public.

Models are read for retention, whether API data trains models (and whether that's off by default), zero-retention options and certifications. Frameworks for telemetry defaults, approval hooks, guardrails and sandboxing.

## 2. Agent ergonomics, 65 out of 100, up to 5.7 more on the total

Why it scored 65: Three tools by default and four with chDB, one of them with a long description (25). `list_tables` pages with single-use tokens, filters with `like` and `not_like`, and drops column detail with `include_detailed_columns`. `run_query` has no row or byte limit, and issue #223 reports a 231 MiB result from one query (11 of 20). ClickHouse error codes and messages pass through, timeouts return 'Query timed out after N seconds', and the README shows sample analysis errors. No documented error list (13 of 20). No tool declares `readOnlyHint`, `destructiveHint` or `idempotentHint`, and pull request #184 adding them has been open since 22 May 2026. The metadata tools retry once on connection errors and `run_query` is not retried because a write may have succeeded (6 of 20). `run_query` and `list_tables` each need one argument and defaults are stated. The server is a Python package only (10 of 15).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-ergonomics):

- 0 to 25, context cost. For MCP, the number and size of the tool definitions (25 for ten or fewer compact tools, 15 for 11 to 30, 5 for more than 30, plus up to 10 back for toolsets, dynamic loading or read-only subsets). For APIs, whether responses can be sized (field selection, limits, summaries).
- 20, pagination, filtering and output-size controls.
- 20, actionable, documented error responses, codes and messages an agent can recover from.
- 20, idempotency or safe retries, and for MCP the `readOnlyHint` and `destructiveHint` annotations.
- 15, sensible defaults, few required parameters, and official SDKs in at least two languages.

Models are read for tool use, structured output, prompt caching, context length, batch and SDKs. Frameworks for how much code and how many defaults a tool-calling agent with MCP needs.

## 3. Reliability, 74 out of 100, up to 5.2 more on the total

Why it scored 74: Scored as local software, since the owner runs it over stdio or its own HTTP listener. Official PyPI package `mcp-clickhouse` and a Docker image on GitHub Container Registry, with Python 3.10 to 3.14 stated (20). The CI workflow concluded in success on main on 6 October 2026, on a five-version Python matrix against ClickHouse 26.3, and the repository holds 403 test functions (25). Three open issues and five open pull requests. Issue #223 on unbounded `run_query` results has been open since 23 August 2026 with no reply, and #155 on per-user access has ten comments (20 of 25). A changelog with a labelled breaking change in 0.5.0 and Compatibility sections in 0.6.0 and 0.7.0, but the changelog file dates only from 14 April 2026, 0.3.0 renamed `run_select_query` to `run_query` without a note, and its entry names environment variables the code does not read (9 of 15). Version 0.7.0, with no stability declaration found (0).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-reliability):

Hosted APIs, MCP servers, models and platforms.

- 20, a public status page with component history (Statuspage, Instatus, BetterStack or the vendor's own).
- 0 to 30, the incident record for the last 90 days on that page. 30 for a clean record or trivial incidents only, 20 for minor incidents only, 10 for one major outage (an hour or more of a core API down, or errors across the board), 0 for several. 5 when there's no history we could read, and the note says so.
- 15, rate limits documented with numbers.
- 15, documented 429 or overload handling (Retry-After, backoff guidance), and idempotency keys or safe-retry guidance where writes are involved.
- 10, an SLA published for any paid tier.
- 10, the surface agents use is generally available, not beta or preview.

Local packages, SDKs, frameworks and stdio MCP servers.

- 20, installs from an official package with supported runtimes stated.
- 25, a public CI and test suite, passing on the default branch.
- 0 to 25, open crash or regression issues relative to activity (25 for few and handled, 0 for many, old and unanswered).
- 15, semver discipline and breaking changes called out in a changelog.
- 15, version 1.0 or later, or declared stable.

Protocols are read from their reference implementations, the public facilitators or servers, spec stability and test vectors.

## 4. Payments & pricing, 60 out of 100, up to 5 more on the total

Why it scored 60: Read with the self-hosted rule. The server is free under Apache-2.0 and needs no signup of its own, so 20 + 20 + 20. No x402, MPP or L402 (0). It needs a ClickHouse database, which can be free open-source ClickHouse, the embedded chDB engine or the public SQL playground the README names. ClickHouse Cloud prices were not checked. The hosted Remote MCP server is a ClickHouse Cloud feature and is not what this listing grades.

The checklist (https://www.anchorterminal.com/benchmark/#checklist-payments):

The published rubric, also on the [x402 page](https://www.anchorterminal.com/x402/).

- 40, a machine payment protocol (x402, MPP or L402) on the tool's own endpoints. 10 to 30 when it covers only some endpoints or only goes through a third party, and the note says which.
- 20, per-call or per-unit pricing published without a login. 10 for public plan-only pricing, 0 for "contact sales" or prices behind a login.
- 20, a free tier or trial that doesn't need a card.
- 20, autonomous onboarding, meaning an agent can get access without a person signing up in a browser (keyless use, x402, a programmatic key API).

Payment platforms and agent wallets rarely charge for their own API over a machine protocol, so the first line has steps for them, and the highest one that applies counts. 40 when x402, MPP or L402 runs on all their own endpoints, 30 when it runs on part of their own API, 25 when their merchants can accept one, 20 for running a facilitator, 15 for paying as a buyer, and 0 when the only protocol is their own. Merchant acceptance sits above a facilitator because the platform's own customers can charge agents through it, while a facilitator settles for sellers who wire up the protocol themselves. The counter-argument (a facilitator does more for the protocol as a whole) has a point. Each note says which step applied.

Open-source software you run yourself is scored on its hosted or paid option if it has one. A free, self-hosted package with nothing to buy gets 20, 20 and 20 for the last three lines, and 0 to 40 for the first only if it ships a payment protocol.

## 5. Schema & documentation, 79 out of 100, up to 3.4 more on the total

Why it scored 79: Each tool is built from a typed Python signature, so inputs carry JSON Schema, read from the source since we do not run vendor software. `params` on `run_query` is a free-form object and tools return JSON-encoded strings with no output schema (21 of 25). clickhouse.com serves llms.txt, a docs llms.txt and a Markdown form of each docs page, and the README is Markdown (10). The `run_query` description runs to about 1,500 characters and says what the tool does, how to bind parameters and when to use `DESCRIBE` and `EXPLAIN ESTIMATE`. `list_tables` documents every argument and the response shape. `list_databases` has one line (16 of 20). `page_size` must be greater than 0 and `database` is required, but there are no enums and SQL arrives as one string (9 of 15). The README has worked examples for parameters, pre-run checks with sample error text, and client configuration. No catalogue of error responses was found (10 of 15). Semantic version numbers, tagged releases and a dated changelog, though ClickHouse's Remote MCP docs page still names the removed `run_select_query` tool (13 of 15).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-schema):

APIs and MCP servers.

- 25, a machine-readable contract (a public OpenAPI file or similar; for MCP, typed JSON Schema inputs on every tool).
- 10, llms.txt or Markdown docs served for agents.
- 0 to 20, descriptions that say what a tool is for, when to use it and when not to, read from the tool definitions in the source or the API reference.
- 0 to 15, typed inputs with enums, constraints and required fields, and no free-form JSON blobs.
- 0 to 15, examples and documented error responses.
- 15, versioning and a public changelog.

Models are read from the API reference, the OpenAPI file, llms.txt, the structured-output and tool-use docs and the model cards. Frameworks from docs a model can follow, typed interfaces, examples and the API reference.

## 6. Transparency & trust, 82 out of 100, up to 1.6 more on the total

Made of editorial 76, provenance 87.

Why it scored 82: Apache-2.0 (30). Local software that connects only to the configured ClickHouse. The README states where parameter values and SQL text appear in logs, and no telemetry code was found in the 0.7.0 source. Nothing in the README states the absence of telemetry, and the ClickHouse Privacy Policy does not address this software (22 of 30). No deprecation policy. The SSE transport is marked deprecated in 0.6.0 with no removal date, and `run_select_query` was removed in 0.3.0 with no prior notice found (8 of 20). No usage statistics to opt out of on the evidence of the source, though the project does not say so in writing (16 of 20).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-transparency):

- 0 to 30, source availability and licence clarity. 30 for open source under an OSI licence, 15 for closed with clear terms, 0 for unclear terms.
- 0 to 30, data handling and retention statements that agree with each other (privacy policy, DPA, retention periods, subprocessors).
- 0 to 20, a deprecation policy or notices with dates.
- 0 to 20, telemetry disclosed with an opt-out (local software), or subprocessors and data locations disclosed (hosted).

The other half of Transparency and trust is the provenance score, computed from checked facts (below). The category score is the mean of the two.

Provenance checks not met in full (half of this category, computed from checked facts):

- Status page: not found (0 of 10)

## 7. Maintenance & community, 94 out of 100, up to 0.5 more on the total

Why it scored 94: v0.7.0 was tagged and released on GitHub on 21 September 2026, 18 days before the check (30). Four tagged releases in the 90 days before the check, 0.4.1 on 17 July, 0.5.0 on 1 September, 0.6.0 on 3 September and 0.7.0 (20). Three open issues against a dozen closed between July and October 2026, most with a maintainer reply and a fix version. Issue #223 has had no reply in six weeks and pull request #184 has been open since May (20 of 25). Listed in the official MCP registry as `io.github.ClickHouse/mcp-clickhouse`, a GitHub-verified namespace, with 0.7.0 published on 21 September 2026 and marked latest (15). Dependabot updates merged on 1 October 2026, a lock file, and a CI matrix that tests the minimum supported driver. `server.json` on main still says 0.6.0 (9 of 10).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-maintenance):

- 0 to 30, time since the last release, or the last published model or API change for a closed service. 30 within 30 days, 20 within 90, 10 within 180, 0 older.
- 20, at least three releases or dated changelog entries in the last 90 days.
- 0 to 25, responsiveness. Issues and pull requests answered on GitHub (the open issues and how recent the replies are). For closed services, a public changelog and a support or community channel that answers, 0 to 15.
- 15, presence in the official MCP registry under a verified namespace (MCP servers), or current official SDKs (APIs and models).
- 10, package health, current dependencies and CI.

Models are read for deprecation notice periods and model churn rather than release counts.

## Deductions

Each comes off the total. A fixed and documented problem counts for less at the next check.

- -3: release 0.3.0 on 14 April 2026 renamed the `run_select_query` tool to `run_query` (commit d82fc87, pull request #93). Neither the changelog entry nor the GitHub release notes for 0.3.0 mention the rename, and ClickHouse's Remote MCP docs page still names the old tool (https://github.com/ClickHouse/mcp-clickhouse/blob/main/CHANGELOG.md).
- -3: issue #131, opened on 18 February 2026, reports that `run_select_query` ran CREATE and DROP statements against a ClickHouse Cloud user with full privileges although the docs said every query ran with `readonly=1`, and that an agent dropped a production table. This is the reporter's account. A maintainer closed it on 20 April 2026 as addressed by pull request #93 in 0.3.0. No advisory was published, and the fix shipped, so a reduced deduction (https://github.com/ClickHouse/mcp-clickhouse/issues/131).
- -2: 0.5.0 on 1 September 2026 closed a `fastmcp run` launch path that served HTTP without authentication and fixed TRUNCATE and `ALTER TABLE ... DROP` forms that the destructive gate had let through with write access on. Both are described in the changelog and fixed, with no advisory, so a reduced deduction (https://github.com/ClickHouse/mcp-clickhouse/blob/main/CHANGELOG.md).

## What we couldn't check

What we couldn't read counted as absent. Publishing it on a page a plain HTTP fetch can read (not only in a browser) lets the next check count it.

- The lead said no hosted server was shown. ClickHouse Cloud has a hosted Remote MCP server at https://mcp.clickhouse.cloud/mcp with OAuth 2.0 or a Cloud API key and 13 read-only tools. It is a separate Cloud feature and this listing grades the open-source server only
- unchecked: the PyPI project page, which answered with a JavaScript client challenge, and the PyPI JSON API, which robots.txt disallows. The 0.7.0 release date rests on the git tag, the GitHub release and the MCP registry entry, all 21 September 2026. The changelog heading says 18 September
- unchecked: tool definitions as a client receives them from `tools/list`. We did not run the server, so input schemas and the absence of annotations are read from the source
- unchecked: the claim in issue #131 that DDL ran under `readonly=1` and that a production table was dropped. It is the reporter's account and no maintainer confirmed or disputed it in the thread
- unchecked: ClickHouse Cloud pricing and its trial, which matter only when the server points at ClickHouse Cloud
- unchecked: the Bugcrowd programme page and whether mcp-clickhouse is in scope
- unchecked: the GitHub Container Registry image page, which rests on the changelog and `server.json`
- No privacy document addresses this software, so `provenance.privacy` and `provenance.terms` are both left out
- The server's MCP instructions tell agents to run `npx -y skills add clickhouse/agent-skills --all`. Recorded as a fact with no deduction, and not acted on

## Weaknesses

- `run_query` returns every row. Issue #223, open since 23 August 2026 with no reply, reports a 231 MiB result from one `SELECT *`
- No tool declares `readOnlyHint` or `destructiveHint`. Pull request #184 adding annotations has been open since 22 May 2026
- Release 0.3.0 renamed `run_select_query` to `run_query` without saying so in its changelog or release notes, and ClickHouse's Remote MCP docs page still uses the old name
- The destructive-statement gate is a keyword check in the server, which the README calls a best-effort guard and not a security boundary
- No prompt-injection guidance for query results was found, the repository has no SECURITY.md, and no advisory covers the security fixes in 0.3.0 and 0.5.0

## What costs an agent a turn today

The notes we give agents before they call it. Each one is a workaround an agent shouldn't need.

- Put a `LIMIT` on every `run_query` call. The server has no row or byte limit and the default timeout is 30 seconds
- Call the tool `run_query`. ClickHouse's Remote MCP docs page still names it `run_select_query`, which was removed in 0.3.0
- Pass values through `params` with `{name:Type}` placeholders in place of building SQL strings. Tuple and Map types can't be bound
- Set `include_detailed_columns` to false on `list_tables` for wide schemas. Page tokens are single-use and expire after one hour
- Connect with a dedicated ClickHouse user holding only the grants needed, and point `CLICKHOUSE_PORT` at the HTTP interface (8123 or 8443), not 9000

## When it's done

Send what changed and where it's published as a dispute (https://www.anchorterminal.com/builders/#disputes, or `POST https://www.anchorterminal.com/api/v1/contact` with `"kind": "dispute"`). Disputes are answered in public, and the listing is checked again by the same checklist. Paying for an audit or a listing claim changes nothing here.