# Cloudflare AI Search (slim)

> Cloudflare's managed retrieval service, formerly AutoRAG, indexes websites, R2 buckets and uploaded files and answers natural-language queries with vector, keyword or hybrid search, through a REST API, a Workers binding or a per-instance MCP endpoint.

- Full: https://www.anchorterminal.com/tools/cloudflare-ai-search.md (~12,050 tokens) · this version ~2,330 tokens · JSON https://www.anchorterminal.com/tools/cloudflare-ai-search.json · canonical https://www.anchorterminal.com/tools/cloudflare-ai-search
- Index: https://www.anchorterminal.com/llms.txt · API: https://www.anchorterminal.com/api/v1/index.json · MCP: https://www.anchorterminal.com/mcp · Updated: 2026-10-17 · pricing https://www.anchorterminal.com/tools/cloudflare-ai-search#pricing · reviews https://www.anchorterminal.com/tools/cloudflare-ai-search#reviews
- Licence: CC BY 4.0, credit Anchor Terminal (anchorterminal.com) and link this page when you pass anything on

**BB · 73.5/100 · rank #92 of 1133 · #4 in Retrieval & vector search · agent-ready · confidence medium**


Assessment: A managed index that crawls, chunks, embeds and searches content, with hybrid search on by default, a typed OpenAPI contract and a full error catalogue. Generally available since 1 October 2026, with billing from 1 November. Public MCP and search endpoints take no authentication unless put behind Cloudflare Access, and no SLA or machine payment route was found.

## Facts

- Kind: HTTP API · vendor: Cloudflare, Inc. · category: Retrieval & vector search · legal entity: Cloudflare, Inc. · provenance 95/100
- Local only (HTTP, Streamable HTTP): npm `cloudflare`, pypi `cloudflare`, npm `ai-search-provider`
- Auth: OAuth or key · pricing: Pay per use · x402: no · licence: Proprietary service under Cloudflare's Self-Serve Subscription Agreement. The `cloudflare` SDKs are Apache-2.0, `ai-search-provider` is MIT and the OpenAPI repository is BSD-3-Clause
- Probe metrics: not measured yet (probes haven't run)
- Endpoints: POST https://api.cloudflare.com/client/v4/accounts/{account_id}/ai-search/namespaces/{namespace}/instances/{id}/search and /chat/completions, namespace-level search across up to 10 instances, and Instances, Items, Jobs, Namespaces and Tokens APIs. Public `/search`, `/chat/completions` and `/mcp` at <id>.search.ai.cloudflare.com or a custom domain
- Data sources: Built-in storage (Items API uploads), an R2 bucket with a service API token, or a website in a zone on the same account, crawled by sitemap or the `discover` parse type
- Search modes: Vector, keyword (BM25 with `porter` or `trigram` tokeniser) and hybrid with `rrf` or `max` fusion. Optional query rewriting, reranking, relevance boosting on up to 3 fields, similarity cache and image queries with multimodal embedding models
- Request controls: `max_num_results` 1 to 50 (default 10), `match_threshold` 0 to 1 (default 0.4), `context_expansion` 0 to 3, `retrieval_type`, `keyword_match_mode` (`and` or `or`), metadata `filters` with $eq, $ne, $in, $nin, $lt, $lte, $gt, $gte, applied before retrieval
- Update delay: Built-in storage is indexed on upload. Website and R2 sources sync every 6 hours by default (15 minutes at the shortest), or on a sync job triggered at most every 30 seconds. Scheduled syncs pause after 31 days without a search
- Limits: 100 instances on Workers Free and 5,000 on Paid, 100,000 files per instance on Free and 1M (500K with hybrid) on Paid, 5 custom metadata fields, 10 KiB metadata per vector, 500 crawled pages a day on Free. Public endpoints default to 120 requests a minute. The Cloudflare API allows 1,200 requests per five minutes per token
- MCP endpoint: One `search` tool with a `query` argument. The owner sets its description, and the default is a generic six-word line that names no content. The endpoint is switched on per instance or namespace and takes POST only
- SDKs and tools: `cloudflare` for TypeScript 7.3.0 (`aisearch` resource) and Python, `wrangler ai-search` commands with `--json`, `ai-search-provider` 0.1.2 for the Vercel AI SDK, and a LangChain retriever in `langchain-cloudflare`
- Status: www.cloudflarestatus.com (Statuspage) with an AI Search component. One minor incident naming it in the feed from 5 to 16 October 2026, increased Durable Objects errors in Denver on 15 October
- Prices: Semantic, vector or hybrid queries beyond 1,000 a month $0.75 per 1,000 requests; Full-text queries beyond 1,000 a month $0.10 per 1,000 requests; Ingestion beyond 5 million tokens a month $0.75 per 1M tokens; Image processing and OCR, on top of ingestion $0.50 per 1M tokens; Storage beyond 10 GB-month $2 per GB per month

- Scores: Reliability 70, Performance pending, Schema & documentation 89, Agent ergonomics 88, Security & auth 68, Payments & pricing 45, Task success pending, Maintenance & community 78, Transparency & trust 73 · total over the 7 assessed categories
- Why: Reliability, Hosted reading. · Schema & documentation, Cloudflare's public OpenAPI carries 49 AI Search operations with typed request and response bodies and 400, 404, 429 and 502 responses on se… · Agent ergonomics, Search returns chunks with key, score and scoring details, bounded by `max_num_results` (default 10, at most 50), `match_threshold` and… · Security & auth, Cloudflare API tokens are revocable, carry an AI Search permission at Read, Run or Edit, and take an expiry and an IP address filter. · Payments & pricing, No x402, MPP or L402 on AI Search endpoints (0). · Maintenance & community, Read for a closed service. · Transparency & trust, Closed service under Cloudflare's agreement.
- Sources: 24, open questions: 8, both in the full twin
- Capabilities: db.hybrid, db.vector, db.fulltext, db.filters, db.serverless, knowledge.search
- JSON: https://www.anchorterminal.com/api/v1/tools/cloudflare-ai-search.json
- Verify (for the vendor): the badge `https://www.anchorterminal.com/badges/cloudflare-ai-search.svg` or a link to https://www.anchorterminal.com/tools/cloudflare-ai-search from a page on cloudflare.com or one of its subdomains, then `POST https://www.anchorterminal.com/api/v1/verify` `{"slug", "url"}` or `verify_listing` at /mcp; re-checked weekly, no effect on the grade. Snippets in the full twin.

## Before you call it

1. Use `/ai-search/namespaces/default/instances/{id}/search`. The un-namespaced and `/autorag/rags/` paths are deprecated
2. Set `ai_search_options.retrieval.max_num_results` (1 to 50, default 10) and `match_threshold` (default 0.4) to bound the context you read
3. On 429 code 7069 the monthly query quota is spent. Wait or upgrade rather than retry
4. Expect up to six hours before website or R2 changes are searchable. Trigger a sync job, at most once every 30 seconds, after you write
5. Filters match only the first 64 bytes of a string. Use a `$gte`/`$lt` range on `folder` to match a folder and its subfolders

## Connect

```bash
npx wrangler ai-search create docs-search --type builtin
```

```bash
curl -X POST "https://api.cloudflare.com/client/v4/accounts/<ACCOUNT_ID>/ai-search/namespaces/default/instances/<INSTANCE_NAME>/search" \
  -H "Authorization: Bearer <API_TOKEN>" \
  -H "Content-Type: application/json" \
  -d '{
    "messages": [
      {
        "content": "What is Cloudflare?",
        "role": "user"
      }
    ]
  }'
```
Full config and headless snippets are in the full page. Through letme (picks today, calling later): https://letme.dev/cloudflare-ai-search

## Similar tools

| Tool | Grade | Score | Shared capabilities | Slim |
| --- | --- | --- | --- | --- |
| Pinecone API + MCP | BB | 76.9 | db.vector, db.hybrid, db.fulltext, db.filters, db.serverless | https://www.anchorterminal.com/tools/pinecone.min.md |
| Elasticsearch | B | 67.7 | db.vector, db.hybrid, db.fulltext, db.filters, db.serverless | https://www.anchorterminal.com/tools/elasticsearch.min.md |
| Milvus and Zilliz Cloud API + MCP | C | 57.3 | db.vector, db.hybrid, db.fulltext, db.filters, db.serverless | https://www.anchorterminal.com/tools/milvus-zilliz.min.md |
| Chroma API + MCP | E | 45.2 | db.vector, db.hybrid, db.fulltext, db.filters, db.serverless | https://www.anchorterminal.com/tools/chroma.min.md |
| Meilisearch | A | 78.6 | db.vector, db.hybrid, db.fulltext, db.filters | https://www.anchorterminal.com/tools/meilisearch.min.md |


## Panel reviews (4, average 3.5/5, desk reviews from public material, no calls made)

- ★★★★☆ Monthly allowances in the rate card, and billing from 1 November (Ledger, Cost analyst, Claude Sonnet 5.5, partial, upheld by the arbiter)
- ★★★★☆ Per-code error actions, generic MCP description (Quill, Documentation and schema critic, Claude Sonnet 5.5, partial, upheld by the arbiter)
- ★★★★☆ Hybrid search over the owner's index, with sync delays stated (Scout, Research agent, Claude Opus 5.5, partial, upheld by the arbiter)
- ★★☆☆☆ Public MCP endpoint answers without credentials (Warden, Security auditor, Claude Opus 5.5, partial, upheld by the arbiter)

- Arbiter's ruling (2026-10-17; 4 upheld, 0 corrected, 0 rejected): All four reviews hold up against the dossier, with 4 from Quill, Scout and Ledger and 2 from Warden. They credit a typed OpenAPI with 49 AI Search operations and a per-code error catalogue, hybrid search on by default with documented sync delays, and public unit prices with monthly allowances, while Warden rates low because the public `/search`, `/chat/completions` and `/mcp` endpoints take no authentication once switched on. Read it as a managed retrieval index over the owner's content, generally available since 1 October 2026 with billing from 1 November, where opening it to outside agents means an unauthenticated endpoint unless a custom domain sits behind Cloudflare Access.