Published 2026-09-22 · Tested 2026-09-22
ClickHouse
A+ClickHouse received 13 PASS votes and passed 5 of five agent surface checks. The clearest finding came from the find the exact limits task.
Three AI models, GPT 5.6 Sol, Claude Opus 5, and DeepSeek v4 Flash, each read ClickHouse’s public documentation independently and attempted five first-hour developer jobs: run the first ClickHouse Cloud query, find the exact limits, recover from Too many parts, authenticate to the Cloud API, use the Python client.
No accounts, API calls, or code execution were used. Every verdict came from public pages and every published quotation passed a live verification check. Preflight substitutions: T3: The Too many parts insert exception replaces HTTP 429 because the ClickHouse query path documents no 429 response. T4: ClickHouse Cloud API key authentication replaces the webhook task because Cloud notifications are delivered only to email, the Cloud UI, and Slack.
Freshness
How rechecks work- Category
- Databases & analytics
- Tested
- Quotes verified
- Surface rechecked
- Not yet rechecked
97.5% · 78/80 · AI Agent Readiness Score · reading 30 pts · surface 50 pts
Agent surface checks · 50/50
The Reading Test
| Task | GPT 5.6 Sol | Opus 5 | DeepSeek v4F | Consensus |
|---|---|---|---|---|
| Run the first ClickHouse Cloud query | PASS | PASS | PASS | PASS |
| Find the exact limits | PASS | PARTIAL | PARTIAL | PARTIAL |
| Recover from Too many parts | PASS | PASS | PASS | PASS |
| Authenticate to the Cloud API | PASS | PASS | PASS | PASS |
| Use the Python client | PASS | PASS | PASS | PASS |
docs platform: Mintlify (unscored) · verified 2026-09-22
What to fix first
This fix could add up to 2 points to the AI Agent Readiness Score. The list ranks each fix by the points it would add. How the ranking works
- 1
Found: The API rate limits sit on a separate page from the usage limits, and the cloud-tiers and knowledge-base pages contradict the usage-limits numbers.
Fix: Gather the service, warehouse, organization, and API key limits into one usage-limits table, and correct the cloud-tiers and knowledge-base numbers to match.
Evidence: clickhouse.com/
docs/ products/ cloud/ guides/ best-practices/ usagelimits (opens in a new tab)
What the docs get right
- Run the first ClickHouse Cloud query: 3 PASS votes. The quick start covers signup, API-key login, service creation, readiness polling, HTTP querying, and a row-returning client query in one flow.
- Recover from Too many parts: 3 PASS votes. One page identifies both triggering settings, gives diagnostic queries, provides corrective paths, warns against raising limits, and defines recovery verification.
- 5 of 5 agent surface checks. Present: llms.txt, llms-full.txt, markdown mirrors, an MCP server, docs AI.
Run the first ClickHouse Cloud query
PASSPASS consensus from 3 PASS.
The quick start covers signup, API-key login, service creation, readiness polling, HTTP querying, and a row-returning client query in one flow. One page carries signup, service creation, provisioning poll, a first query with its JSON output, and the native-client alternative, with copyable commands at every step. One page with GUI and CLI tabs, each giving a complete signup-to-query path; the CLI tab's first clickhousectl cloud service query --query SHOW databases returns rows immediately after provisioning.
Find the exact limits
PARTIALPARTIAL consensus from 1 PASS, 2 PARTIAL.
The pages consistently state 1,000 concurrent queries per replica, 50 replicas per warehouse, 20 services per organization, and 10 API requests per key per 10 seconds. Service, warehouse, organization, and API rate limits are all stated with exact numbers, but a knowledge-base page gives 100000 parts per service where the usage-limits and cloud-compatibility pages both give 10k. Service, warehouse, and organization limits are documented with specific numbers, and Cloud API rate limits (100 keys/org, 10 req/10s per key) appear on a separate page. However, the cloud-tiers page states Unlimited services for every tier, contradicting the 20-soft-limit on the usage-limits page.
Recover from Too many parts
PASSPASS consensus from 3 PASS.
One page identifies both triggering settings, gives diagnostic queries, provides corrective paths, warns against raising limits, and defines recovery verification. The page names both triggering settings, supplies diagnostic SQL against system.parts, gives four ordered fixes, warns against raising the limits as the primary fix, and closes with a verification step. The KB article identifies the root cause (parts_to_throw_insert / max_parts_in_total), provides diagnostic queries, gives three concrete recovery strategies (batch inserts, async inserts, partitioning key review), warns against increasing limits, and closes with a verify-recovery step.
Authenticate to the Cloud API
PASSPASS consensus from 3 PASS.
The key guide covers role-based permissions, optional IP scope, one-time credential capture, Basic authentication, and an authorized organizations request. Numbered steps run from opening the API Keys page through role scoping, IP allowlisting, and a working curl call against https://api.clickhouse.cloud/v1/organizations using HTTP Basic auth. Seven-step flow from opening the API Keys page to an authenticated curl request, including key scoping via permissions and predefined roles (developer vs admin).
Use the Python client
PASSPASS consensus from 3 PASS.
The Python introduction gives the full port-8443 query path and matches the HTTP docs; the guessed /get-started/quickstarts/use-clickhouse-connect route returned 404. The client page and the Cloud knowledge-base example use the same API surface, get_client(host, port=8443, username, password), command, insert, query, result_rows, so the two agree despite the knowledge-base page pinning a much older driver version. The official client is identified (pip install clickhouse-connect), and code examples show Cloud connection on port 8443 with get_client, command, insert, and query methods. Port 8443 and HTTPS match the HTTP interface docs and Cloud quick start endpoints.
The receipt
Each API key has a limit of 10 requests over a 10-second window.
Service, warehouse, organization, and API rate limits are all stated with exact numbers, but a knowledge-base page gives 100000 parts per service where the usage-limits and cloud-compatibility pages both give 10k.
Agent surface notes
Initialize returned JSON-RPC protocol 2025-06-18 with serverInfo name ClickHouse Documentation.
The hydrated docs home shows a Mintlify Ask AI button with id ch-ask-ai-btn and an assistant input bar, alongside Inkeep and Kapa widget containers.
Show the score
Paste this into a readme:
[](https://docsforagents.com/reports/clickhouse-docs-ai-agent-readiness/) Method note
This is a reading test of public documentation, not an execution test. No accounts were created and no API calls were run. The AI Agent Readiness Score counts fifteen reading votes at PASS 2, PARTIAL 1, and FAIL 0, for 30 possible points. Five agent surface checks add 10 points each. The total is 80. Consensus chips show each row majority and do not affect scoring. The panel split on 1 of five tasks. Quotes shown here were re-fetched and confirmed verbatim on 2026-09-22.
Put another docs site through the battery.
Nominate a docs site