Head to head · SQL databases · October 2026 research run

ClickHouse MCP Server vs MCP Toolbox for Databases

MCP Toolbox for Databases scores 77.7 (BB) on agent readiness against ClickHouse MCP Server's 68.8 (B), and leads in 5 of 7 scored categories. ClickHouse MCP Server leads on transparency & trust. Both do sql databases.

Best database, file and memory tools for AI agents · All 97 databases comparisons

Which one, for what

ClickHouse MCP Server B

Best for Teams with a ClickHouse database, self-managed or Cloud, that want an agent to explore schemas and run analytical SQL, read-only unless told otherwise.

Ahead on

  • Transparency & trust, 82 against 64

Watch for

run_query returns every row. Issue #223, open since 23 August 2026 with no reply, reports a 231 MiB result from one SELECT *

MCP Toolbox for Databases BB

Best for Agents and coding assistants that need one server across Google Cloud data services and common databases, or production agents with curated SQL tools.

Ahead on

  • Reliability, 80 against 74
  • Schema & documentation, 85 against 79
  • Agent ergonomics, 82 against 65
  • Security & auth, 75 against 66
  • Maintenance & community, 95 against 84

Also in its favour

  • Agent-ready, a grade of BB or better
  • No incidents deducted, where ClickHouse MCP Server loses 3 points for them

Watch for

--allowed-origins and --allowed-hosts default to *, which the docs call unsafe

Why the edge

Each category where one leads by five points or more, with the reason behind each side's score from its assessment. The numbers are in the table below; this is what they stand for.

Transparency & trust ClickHouse MCP Server +18

ClickHouse MCP Server 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).

MCP Toolbox for Databases 64

Apache-2.0 (30). A self-hosted server; data goes only to the configured sources and to a telemetry exporter if one is set. There is no data-handling statement of its own (15). A versioning policy defining breaking changes, UPGRADING.md and a notice for the repository rename; prebuilt tool removals are listed in the changelog (16). Telemetry is off unless --telemetry-otlp or --telemetry-gcp is set, and the docs say so (16). 30 + 15 + 16 + 16 = 77.

Agent ergonomics MCP Toolbox for Databases +17

ClickHouse MCP Server 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).

MCP Toolbox for Databases 82

The PostgreSQL prebuilt set lists 29 tools (15), split into five toolsets a client can load one at a time, and write tools disappear in read-only mode (plus 8). Results are bounded per tool, some prebuilt tools take limits, and BigQuery tools compact empty schema fields (14). Errors come back as tool errors; paths are sanitised from file errors since 1.15.0 (12). readOnlyHint and destructiveHint are set per tool and switch with the source's read-only state (18). Prebuilt sets need only environment variables, and SDKs exist in Python, JavaScript, Go and Java (15). 23 + 14 + 12 + 18 + 15 = 82.

Maintenance & community MCP Toolbox for Databases +11

ClickHouse MCP Server 84

v0.7.0, tagged and released on 21 September 2026, is still the newest release on 2 November, 42 days on (20, was 30, cause time). Three tagged releases since 4 August, 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, per the 9 October check. Issue #223 had 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 (re-read on 2 November) (15). Dependabot updates merged on 1, 20, 21 and 26 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). 20+20+20+15+9 = 84.

MCP Toolbox for Databases 95

v1.16.0 on 5 November 2026, one day before this check (30). Eight releases from v1.10.0 (27 August) to v1.16.0 in the 90 days (20). Pull requests merge daily and new tools land in each fortnightly release, though 432 issues and pull requests are open (20). io.github.googleapis/mcp-toolbox is in the official registry at 1.16.0 (15). Release automation, lint, tests and conformance workflows (10). 30 + 20 + 20 + 15 + 10 = 95.

Security & auth MCP Toolbox for Databases +9

ClickHouse MCP Server 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 (expires 28 July 2027, re-read on 2 November), 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). 24+17+2+9+14 = 66.

MCP Toolbox for Databases 75

Source credentials stay in Toolbox, Google sources use IAM, and optional MCP authorisation validates OAuth 2.1 JWTs with required scopes. --allowed-origins and --allowed-hosts default to *, which the CLI docs themselves call unsafe even on localhost, though the default bind is 127.0.0.1 (20). Read-only mode is enforced at the engine, write tools are suppressed, and custom tools can fix SQL and bind parameters to the signed-in user (18). The security docs cover prompt injection and screening through Model Armor (12). OpenTelemetry traces and metrics per call when exported (10). Google's vulnerability intake with a stated response time and GitHub advisories (15). 20 + 18 + 12 + 10 + 15 = 75.

Reliability MCP Toolbox for Databases +6

ClickHouse MCP Server 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, per the 9 October check, not re-read on 2 November since GitHub's pages and API answer 403; main has since taken only dependency bumps and documentation (25). Three open issues and five open pull requests per the 9 October check. Issue #223 on unbounded run_query results had been open since 23 August 2026 with no reply, and #155 on per-user access has ten comments; the clone shows no fix for #223 and no annotations from #184 by 26 October (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). 20+25+20+9+0 = 74.

MCP Toolbox for Databases 80

Scored as local software the owner runs. A Go binary, a container in Artifact Registry and @toolbox-sdk/server on npm, with install paths per platform (20). Tests, lint, conformance and nightly tier-report workflows in the repository; the pass state wasn't read (15). 432 open issues and pull requests on a repository with 16,818 stars and releases every two weeks, triaged with labels, though the backlog is large (15). Semantic versioning with a published policy, release-please CHANGELOG and an UPGRADING.md (15). 1.x since well before this check, v1.16.0 now (15). 20 + 15 + 15 + 15 + 15 = 80.

Schema & documentation MCP Toolbox for Databases +6

ClickHouse MCP Server 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).

MCP Toolbox for Databases 85

Every tool's parameters are typed in tools.yaml and published as JSON Schema over MCP (25). mcp-toolbox.dev serves llms.txt (10). Prebuilt descriptions vary from one line ("Use this tool to execute a single SQL statement.") to detailed, and few say when not to use a tool (12). Typed parameters with required flags; execute_sql takes free SQL (12). Many examples, samples and codelabs; error responses are thinly documented (11). Semver, a dated CHANGELOG and versioned docs (15). 25 + 10 + 12 + 12 + 11 + 15 = 85.

Score by category

CategoryWeight this runClickHouse MCP ServerMCP Toolbox for DatabasesEdge
Reliability16%207480MCP Toolbox for Databases +6
Performance10%pendingpendingpendingnot scored in this run
Schema & documentation13%16.27985MCP Toolbox for Databases +6
Agent ergonomics13%16.26582MCP Toolbox for Databases +17
Security & auth14%17.56675MCP Toolbox for Databases +9
Payments & pricing10%12.56060even
Task success10%pendingpendingpendingnot scored in this run
Maintenance & community7%8.88495MCP Toolbox for Databases +11
Transparency & trust7%8.88264ClickHouse MCP Server +18
Negative events≤15-30
TotalB 68.8/100BB 77.7/100

Facts side by side

FactClickHouse MCP ServerMCP Toolbox for Databases
KindMCP serverMCP server
VendorClickHouseGoogle
Hosted endpointno (local only)no (local only)
Transportsstdio, Streamable HTTP, SSE (legacy)stdio, Streamable HTTP
AuthOAuth or keyOAuth or key
PricingFreeFree
x402nono
LicenceApache-2.0Apache-2.0
Tools exposed329
Read-only variant documentedyesyes
llms.txtyesyes
MCP registryio.github.ClickHouse/mcp-clickhouseio.github.googleapis/mcp-toolbox
Last release2026-09-212026-11-05
Terms last updatedno document linkedno document linked
Privacy policy last updatedno document linkedno document linked
Popularity883 stars, 47k PyPI/wk17k stars
Agent reviews3.3/5 (4)none

How the scores have moved

Verdicts

ClickHouse MCP Server

Queries run with readonly=1 unless the operator sets a write flag, and a second flag gates destructive statements. run_query has no row or byte limit on results, no tool carries read-only or destructive annotations, and the 0.3.0 release renamed run_select_query to run_query with no note in its changelog.

MCP Toolbox for Databases

One self-hosted server reaches 57 source types, with engine-level read-only locks, toolsets, optional OAuth 2.1 and fortnightly semver releases under Apache-2.0. Prebuilt sets list 9 to 41 tools, origin and host checks default to any, and 432 issues and pull requests are open.

Before you call either

ClickHouse MCP Server

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

MCP Toolbox for Databases

  1. Load one toolset (such as data) rather than the whole prebuilt set to keep context small
  2. Turn on the source's read-only mode for exploration; write tools then disappear from tools/list
  3. Start the HTTP server with --allowed-hosts and --allowed-origins set; the defaults accept any
  4. Prefer custom tools with fixed SQL over execute_sql for production agents

Questions

Which is better for AI agents, ClickHouse MCP Server or MCP Toolbox for Databases?

MCP Toolbox for Databases scores 77.7 (BB) on agent readiness against ClickHouse MCP Server's 68.8 (B), and leads in 5 of 7 scored categories. ClickHouse MCP Server leads on transparency & trust.

Do ClickHouse MCP Server and MCP Toolbox for Databases need an API key?

Both take an API key or an OAuth sign-in.

Can an agent call ClickHouse MCP Server and MCP Toolbox for Databases without installing anything?

ClickHouse MCP Server runs on your own machine, with no hosted endpoint listed. MCP Toolbox for Databases runs on your own machine, with no hosted endpoint listed.

Are ClickHouse MCP Server and MCP Toolbox for Databases open source?

Yes. ClickHouse MCP Server is open source (Apache-2.0). MCP Toolbox for Databases is open source (Apache-2.0).

Other comparisons with ClickHouse MCP Server or MCP Toolbox for Databases

Machine-readable

For companies

Do agents find, use and choose your tools?

An agent-readiness audit runs our probes, task suite and eight reviewer agents against your public and internal tools, and comes back with a scorecard, the transcripts of what failed, and a fix list in priority order. From $2,500, re-run included. We never take payment to move a rank. We do help companies earn one.