Milvus and Zilliz Cloud API + MCP by Zilliz

# Fix list: Milvus and Zilliz Cloud API + MCP

From Anchor Terminal's listing at https://www.anchorterminal.com/tools/milvus-zilliz, the October 2026 research run, assessed 1 October 2026. Grade C, 57.5 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 Milvus and Zilliz Cloud API + MCP: 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, 53 out of 100, up to 8.2 more on the total

Why it scored 53: Zilliz Cloud has personal keys and customised keys scoped to projects and clusters with a role, which owners can reset. Cluster username and password also work as a token. Self-hosted Milvus ships with authentication off (28). We found no read-only role in the key docs, and the MCP server has no read-only mode, though it can delete entities and suspend clusters (8). Search returns stored data as written and we found no prompt-injection guidance. Until 18 August 2026 the MCP data tools took the data-plane URL from the caller (5). No audit log found in the docs we read (0). SOC 2 Type II and ISO 27001 on the pricing page, HIPAA ready on Business Critical and a trust centre page. No SECURITY.md in the Milvus repo and no security.txt per the 30 September check, but two critical advisories were published with patched versions on every supported branch (12).

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. Payments & pricing, 35 out of 100, up to 8.1 more on the total

Why it scored 35: No x402, MPP or L402 (0). Plans are public and the docs give $4 per million vCUs, but the pricing page itself sends Serverless and Dedicated rates to a calculator (15). Free cluster with 5 GB and 2.5M vCUs a month, no card per the 30 September check (20). A person signs up in the browser. The MCP server can create a free cluster, but only with a token from a signed-up account (0).

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.

## 3. Agent ergonomics, 62 out of 100, up to 6.2 more on the total

Why it scored 62: 16 MCP tools, 7 for the control plane and 9 for data (15). `limit`, `offset`, `output_fields`, filter expressions, grouping and partition selection on search and query (20). Milvus error codes are documented, but MCP errors are pass-through text (12). The API has upsert, but the MCP server only exposes insert, and no tool carries readOnlyHint or destructiveHint, including `delete_entities` and `suspend_cluster` (5). Official clients in Python, Node, Go and Java. Every MCP data call needs `cluster_id` and `region_id`, and a collection needs a schema and an index before search (10).

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.

## 4. Schema & documentation, 69 out of 100, up to 5 more on the total

Why it scored 69: No OpenAPI file found for the Zilliz Cloud REST API, though the gRPC protocol is public and the MCP tools carry typed Python signatures (15 of 25). llms.txt on milvus.io and docs.zilliz.com, with the REST reference as a text file for agents (10). MCP docstrings state purpose and explain each argument, but none says when not to use a tool (10). Filters are a string expression and `search_params` a free-form dict, with no enums for metric or consistency level (9). Examples throughout the docs and documented Milvus error codes, but the MCP server passes back "Business error" plus the raw message (10). Dated Zilliz changelog and Milvus release notes with semver (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.

## 5. Reliability, 85 out of 100, up to 3 more on the total

Why it scored 85: Atlassian Statuspage at status.zilliz.com with an RSS history back to December 2023 (20). One incident since July, 1 hour 7 minutes of authorisation errors on a subset of control-plane requests on 22 July 2026, with running clusters and data access unaffected, so minor only (20). Limits published with numbers, 5 collection operations a second per cluster, insert throughput from 2 MB/s on Free to 64 MB/s on large Dedicated clusters, 10 query vectors and 1,024 results a search on Free and Serverless (15). The limits page gives no error code or retry advice, but pymilvus retries rate-limit errors with backoff by default, and upserts by primary key are safe to repeat (10 of 15). 99.95% SLA on Dedicated Enterprise and 99.99% on Business Critical with replicas (10). GA (10).

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.

## 6. Transparency & trust, 70 out of 100, up to 2.6 more on the total

Made of editorial 65, provenance 75.

Why it scored 70: Milvus is Apache-2.0 under the LF AI and Data Foundation (30). The Zilliz privacy policy, updated July 2025, names Zilliz Inc. and some recipients, but states no retention period, links no DPA and no subprocessor list (15). No deprecation or end-of-life policy found beyond release notes (5). Open-source Milvus sends no telemetry to Zilliz that we could find. Its 2026 client telemetry reports to the operator's own server, as described in a public design doc (15).

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):

- Endpoint on the vendor's domain: api.cloud.zilliz.com is not on zilliz.com (0 of 15)
- security.txt: not found (0 of 10)

## 7. Maintenance & community, 85 out of 100, up to 1.3 more on the total

Why it scored 85: Milvus v2.6.25 tagged on 28 September and v3.0.2 on 18 September 2026 (30). v3.0.0, v3.0.1, v3.0.2 and seven 2.6 patch releases since July (20). 1,100 open issues with kind and difficulty labels on new ones, few assignees visible (15). pymilvus 3.0.2 and the Node SDK 3.0.6 follow the 3.0 server (15). 21 CI workflows including code checks, licence checks and weekly releases (10). Less 5 because the Zilliz MCP server's only PyPI release is 1.0.0 from 30 June 2025, which predates the August security fix.

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.

- CVE-2025-64513 (GHSA-mhjq-8c7m-3f7p), CVSS 9.3, published 10 November 2025, an authentication bypass in the Milvus Proxy giving full admin access. Patched in 2.4.24, 2.5.21 and 2.6.5. Fixed and disclosed, so a small deduction (-2) (https://github.com/advisories/GHSA-mhjq-8c7m-3f7p).
- CVE-2026-26190 (GHSA-7ppg-37fh-vcr6), CVSS 9.8, published 11 February 2026, unauthenticated access to REST APIs on metrics port 9091. Patched in 2.5.27 and 2.6.10. Fixed and disclosed (-2) (https://github.com/advisories/GHSA-7ppg-37fh-vcr6).
- Before 18 August 2026 the Zilliz MCP data tools accepted a caller-supplied endpoint, so a prompt-injected URL could receive the server-held ZILLIZ_CLOUD_TOKEN. The fix is on GitHub, but the README still installs with `uvx zilliz-mcp-server` and the only PyPI release is 1.0.0 from 30 June 2025, with no advisory (-4) (https://github.com/zilliztech/zilliz-mcp-server/commit/23beec692734e61c688e57fc8a8bf8f4cf610ab5, https://pypi.org/project/zilliz-mcp-server/#history).

## 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.

- Whether the PyPI build of zilliz-mcp-server 1.0.0 takes the endpoint from the caller like the pre-fix GitHub code. We read the source, not the wheel
- unchecked: whether Zilliz Cloud has audit logs or a read-only key role
- Whether Zilliz publishes an OpenAPI file for its v2 REST API

## Weaknesses

- The Zilliz MCP server's only PyPI release dates from June 2025 and predates an August 2026 fix for token forwarding
- Two critical Milvus advisories in the past year, CVE-2025-64513 and CVE-2026-26190, both patched
- 16 MCP tools with no annotations or read-only mode, including delete_entities and suspend_cluster
- No OpenAPI file for the Zilliz Cloud REST API
- Collections need a schema and an index before search, and every MCP data call needs cluster and region IDs

## 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.

- Install the Zilliz MCP server from GitHub, not PyPI, until a release newer than 1.0.0 ships
- Use the cluster endpoint for data (`/v2/vectordb/entities/search`) and api.cloud.zilliz.com only for cluster management
- Set `consistencyLevel` to Strong when a test needs a write visible to the next query. The default is Bounded
- Expect at most 1,024 results and 10 query vectors per search on Free and Serverless
- Return only the fields you need. Returning the vector field multiplies read cost

## What the review panel asked for

- a 2.6 end-of-life date
- a fixed PyPI release
- Put rates on the page

## 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.