Snowflake-managed MCP server

# Fix list: Snowflake-managed MCP server

From Anchor Terminal's listing at https://www.anchorterminal.com/tools/snowflake-managed, the October 2026 research run, assessed 9 October 2026. Grade C, 56.4 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 Snowflake-managed 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. Reliability, 55 out of 100, up to 9 more on the total

Why it scored 55: Graded as a hosted service. Statuspage at status.snowflake.com with components by cloud region (20). Only the front page's 15 days could be read. They show two incidents marked critical, core services unavailable in one region for 68 minutes on 25 September 2026 and in other regions for 35 minutes on 27 September, one marked major (Snowsight connectivity, about two and a half hours on 6 October) and one minor incident open on 8 October. All were regional. One major outage, with older history unread (10). No rate limit for the MCP endpoint found (0). The REST API's OpenAPI files describe 429 and retry with backoff, but the MCP page gives no retry or idempotency guidance (5). 99.9 per cent monthly SLA with service credits (10). Generally available since 4 November 2025 (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.

## 2. Agent ergonomics, 46 out of 100, up to 8.8 more on the total

Why it scored 46: The tool list is whatever the operator declares, up to 50 a server, and Snowflake recommends one Cortex Agent tool per server with further servers for other uses (18). The search tool takes `columns` and `limit`. SQL and custom tool responses are cut at 250 KB with no paging, and agent tool responses carry every intermediate step and can pass 200 KB (8). Errors are documented as a symptom table, and `invalid_client` covers both wrong credentials and a network policy block (6). The SQL tool is read-only by default. No `readOnlyHint` or `destructiveHint` annotations and no idempotency support are documented (6). Few required arguments and standard MCP clients work with no SDK, but each client needs a hand-made OAuth integration and some clients can't choose a role (8).

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

Why it scored 35: No x402, MPP or L402 (0). Per-credit prices by region and edition and per-unit AI rates are public in the Service Consumption Table of 2 October 2026, though the MCP server has no line of its own (20). A 30-day trial needs only an email address, and Cortex tools stay off until a card is added (15). A person signs up in a browser and an administrator creates the server and the OAuth integration (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.

## 4. Schema & documentation, 61 out of 100, up to 6.3 more on the total

Why it scored 61: No OpenAPI or published schema covers the MCP endpoint, and the snowflake-rest-api-specs repository has no MCP server file. Custom tools carry a JSON Schema the operator writes in `input_schema`, and built-in tools take fixed arguments shown only as examples. The schemas `tools/list` returns were not read (15). llms.txt and a Markdown copy of every docs page (10). Tool names and descriptions are written by the operator, and Snowflake's own examples are one line each, such as "cortex search service for all products" (8). Typed inputs for custom tools, with free-text SQL and a single `message` string for the analyst and agent tools (8). Request and response examples for two tool types, a troubleshooting table and OAuth failure modes, with no JSON-RPC error codes and an empty object as the search response example (8). Dated release notes, behaviour change notes and a stated protocol revision of 2025-11-25 (12).

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. Security & auth, 77 out of 100, up to 4 more on the total

Why it scored 77: OAuth 2.0 with PKCE, role scopes, an allowed-roles list, binding to an external identity provider and RFC 9728 metadata. Programmatic access tokens expire (15 days by default, 365 at most), can be rotated and need a network policy by default. No dynamic client registration (28). Each tool needs its own grant, the SQL tool defaults to read-only, and the docs advise a separate least-privileged server for direct SQL. No confirmation step for writes is documented (15). The docs warn about tool poisoning, tool shadowing and recursive loops and cap recursion at 10. Nothing covers instructions arriving in query or search results (6). Account Usage has query, login and access history and a Cortex Agent usage view. No log of MCP tool calls was found (10). security.txt valid to 17 August 2027 with a HackerOne policy, SOC 2 Type II, ISO 27001, HITRUST and FedRAMP per the Security Addendum. The HackerOne page was not read (18).

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.

## 6. Maintenance & community, 55 out of 100, up to 3.9 more on the total

Why it scored 55: The last dated change to the MCP server is the SSE note of 20 August 2026, 50 days before the check. Platform release notes run to 9 October (20). Three dated MCP entries in 90 days, on 22 July, 7 August and 20 August 2026 (20). A public changelog and a support route through cases and the community, whose answers were not sampled (10). The managed server is not in the official MCP registry, which lists only the separate open-source `io.github.Snowflake-Labs/mcp` (0). Nothing to install. The REST API specification repository was last updated on 10 September 2026 (5).

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.

## 7. Transparency & trust, 84 out of 100, up to 1.4 more on the total

Made of editorial 68, provenance 99.

Why it scored 84: Closed service with clear published terms, an acceptable use policy and a consumption table (15). A DPA and Security Addendum of 11 January 2026 with 72-hour incident notice and customer-chosen regions. The DPA says leftover data is deleted promptly after termination without a day count, and the Privacy Notice excludes Customer Data (22). A written behaviour change policy with an eight-week bundle period and dated notes, though the August 2026 MCP change shipped outside it (15). Four sub-processors and 28 affiliates listed with locations and 28 days' notice of additions. AI model providers are not named on the list (16).

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):

- Terms of service: read, states 6 of the 7 things a reader expects (9.1 of 10)

## Deductions

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

- 20 August 2026. `tools/call` responses changed from a single JSON body to an SSE stream as an unbundled behaviour change that accounts could not disable, outside the eight-week bundle process. Clients reading one JSON body broke. The change is documented with migration steps and clients that follow the MCP specification were unaffected, so the deduction is small. Whether notice was given before the release date was not established (https://docs.snowflake.com/en/release-notes/bcr-bundles/un-bundled/bcr-2405).

## 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.

- unchecked: status history before 25 September 2026. The history page is drawn by script and robots.txt closes the status API, so only the 15 days on the front page were read.
- unchecked: the JSON Schemas that `tools/list` returns for the built-in tool types. No account was used, and the docs show only example arguments.
- unchecked: whether the 20 August 2026 SSE change was announced before it shipped. The note gives only the release date.
- unchecked: the HackerOne programme page named in security.txt, so whether it pays bounties is not established.
- unchecked: the trial's free usage amount. The trial page gives 30 days and no figure.
- No rate limit for the MCP endpoint was found in the reviewed documentation.
- No per-call log of MCP tool calls was found. The Account Usage index lists QUERY_HISTORY, LOGIN_HISTORY, ACCESS_HISTORY and CORTEX_AGENT_USAGE_HISTORY, none specific to MCP.
- Whether a directly exposed Cortex Analyst tool bills at the Analyst API rate (67 Platform Credits per 1,000 messages) is not stated.
- The Privacy Notice says it does not apply to Customer Data, which the customer agreement and DPA govern. `provenance.privacy` points at the notice because it is the only privacy policy Snowflake publishes.
- The Acceptable Use Policy allows benchmark tests only if results are shared with Snowflake before disclosure and can be replicated. This matters before any probe is run.
- The lead was right on the endpoint, the tool types and OAuth. It did not mention programmatic access tokens, which the server also accepts.

## Weaknesses

- No rate limits for the MCP endpoint and no JSON-RPC error codes were found in the reviewed documentation
- No dynamic client registration. An administrator creates a security integration and hands the client ID and secret to each MCP client
- On 20 August 2026 `tools/call` responses changed from one JSON body to an SSE stream, outside a behaviour change bundle and with no way to disable it
- Resources, prompts, roots, notifications, sampling, version negotiation and lifecycle phases are unsupported, and SQL and custom tool responses are cut at 250 KB
- status.snowflake.com shows two incidents marked critical (25 and 27 September 2026) and one marked major (6 October) in the 15 days its front page lists
- Cortex tools stay disabled on a trial account until a credit card is added

## 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.

- Send `Accept: application/json, text/event-stream` and read `tools/call` results as an SSE stream that ends with `data: [DONE]`
- Write the account hostname with hyphens, not underscores, or some clients fail to connect
- Check the user's `DEFAULT_ROLE` and `DEFAULT_WAREHOUSE` first. Clients that request `session:role:all` get the default role, and a missing warehouse stops the session starting
- Keep SQL results narrow. Responses over 250 KB are truncated with no paging, and agent tool responses can pass 200 KB
- Treat `invalid_client` from `/oauth/token-request` as a possible network policy block as well as a wrong credential

## 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.