LangWatch

# Fix list: LangWatch

From Anchor Terminal's listing at https://www.anchorterminal.com/tools/langwatch, the October 2026 research run, assessed 9 October 2026. Grade B, 65.5 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 LangWatch: 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, 53 out of 100, up to 9.4 more on the total

Why it scored 53: Graded as a hosted service (LangWatch Cloud). Better Stack status page with 9 components and 90 days of history (20). In that window the App had one 20-minute downtime on 19 July and the Collector 23 minutes that day and 5 minutes on 13 August, but the Processor was degraded on 17 days, for 14 hours 44 minutes on 31 July, and Scenarios was down on 7 days, for 9 hours 34 minutes on 8 September. The core API stayed under an hour and other components had several long incidents (10). No request rate limit for the platform API was found in the reviewed documentation. The only numbers are customer-set limits on AI Gateway virtual keys (0). Retries are partly covered. Events carry an idempotency key and are deduplicated, `evaluation_id` makes an evaluation call idempotent, and two operations declare a 429, one with `Retry-After`. No general backoff guidance for the platform API was found (8). The pricing page lists a contractual uptime and support SLA on Enterprise, with no published terms (5). The REST API, MCP server 2.1.0 and the 1.x SDKs are generally available (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. Payments & pricing, 40 out of 100, up to 7.5 more on the total

Why it scored 40: No x402, MPP or L402 (0). Per-unit prices are public without a login, 29 euros per core seat a month, 5 euros per 100,000 events and $0.0546 per million judge input tokens (20). The Developer plan is free with no card (20). A person signs up in a browser and creates the first key. The CLI's device-code login also needs a browser approval, and `platform_create_project` returns a key only to a caller that already holds an organisation key. Free self-hosting is not an agent route (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, 67 out of 100, up to 5.4 more on the total

Why it scored 67: The MCP server registers 101 tools in the 2.1.0 source, with no toolsets or dynamic loading found. Trace tools return a compact digest by default, which earns 3 back (8). Cursor paging with `next_cursor`, `limit` and `pageSize`, filters, a projection DSL on trace search and a `digest` or `json` format switch (20). Gateway errors carry a stable code, tips, a docs link and a `fault` field, and the MCP client passes them on. Platform errors are plainer (16). No `readOnlyHint` or `destructiveHint` in the MCP source. Idempotency exists for evaluations and ingested events, and several delete tools archive instead of removing (8). Python, TypeScript and Go SDKs, a CLI, and few required parameters (15).

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. Security & auth, 71 out of 100, up to 5.1 more on the total

Why it scored 71: OAuth authorisation code with PKCE and dynamic client registration on the remote MCP server, and API keys with read or write per category, a project, team or organisation scope, expiry and revocation, plus write-only ingestion keys. The `apiKey` query parameter on `/sse` was removed in MCP server 2.0.0 (30). Restricted keys and lite members give read-only access, but no tool asks for confirmation before a delete and the MCP tools carry no annotations (14). Traces hold whatever the application logged, and we found no prompt-injection guidance for agents reading them (3). The audit log is Enterprise only and leaves out sign-ins and reads of product data. The key list shows a last-used date (9). SECURITY.md sets response targets and a safe harbour, one advisory was published in August 2026, and CodeQL runs in CI. ISO 27001 and a CREST penetration test are stated, with reports on request. No security.txt and no bug bounty found (15).

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.

## 5. Transparency & trust, 75 out of 100, up to 2.2 more on the total

Made of editorial 73, provenance 77.

Why it scored 75: Apache 2.0 platform with the `platform/app/ee` directory under a separate Enterprise licence, and MIT SDKs and MCP server (27). The privacy policy of 29 September 2026 names its processors and the retention docs are detailed, but the documents disagree in places. The pricing page says 30-day retention is included where the docs say 49 days, the terms describe a 14-day free trial where the pricing page says free forever, and the terms give two addresses and two spellings of the entity. The terms also grant LangWatch a perpetual licence to use client data to improve the platform, while saying it will not train on it (18). Deprecated endpoints are marked in the OpenAPI document and the docs, and breaking changes are flagged in the changelogs, but no removal dates or notice period were found (10). EU hosting on AWS is stated and processors are named in the privacy policy. Self-hosted telemetry is documented connection by connection with an off switch. The public sub-processor list is on a host we could not read (18).

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):

- Domain age: langwatch.ai, registered 2023-09-17 (3 years) (7 of 15)
- Terms of service: read, states 7 of the 7 things a reader expects, and has 2 clauses that cost points (6 of 10)
- Privacy policy: read, states 7 of the 8 things a reader expects (9.3 of 10)
- security.txt: not found (0 of 10)

## 6. Schema & documentation, 87 out of 100, up to 2.1 more on the total

Why it scored 87: A public OpenAPI 3.1 document of 215 paths and 328 operations, served without a key, and typed zod inputs on every MCP tool (25). llms.txt and a Markdown twin of each docs page (10). Every operation has a summary or description, and MCP descriptions say when to pick a tool, for example `run_query` for counts and rates and `discover_schema` before a search. 158 operations carry no tag (16). 540 enums in the document, with trace filters still accepted as a free-form map alongside the query language (12). Docs pages carry request examples, and the gateway error list gives 34 codes with cause and remedy. Most platform operations document errors only as an `error` and `message` string pair (11). Dated public changelog, release notes per package and a document version of 1.0.0, with `/api/v1` paths mixed among unversioned ones (13).

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.

## 7. Maintenance & community, 90 out of 100, up to 0.9 more on the total

Why it scored 90: Platform 3.20.1 on 2 October 2026, seven days before the check (30). 21 platform releases dated since 11 July in CHANGELOG.md (20). 52 open issues and 222 open pull requests, with new issues filed, labelled and prioritised by maintainers within the past week, and support by email, Discord and GitHub. Reply times were not visible to us (17). Current official SDKs, Python 1.4.0 on 6 September, TypeScript 1.19.0 on 30 September and Go 1.0.0 on 9 September. The MCP server was not found in the official MCP registry (15). CI workflows for the app, each SDK and the MCP server, CodeQL, code scanners and dependency bumps in the changelog. We could not confirm the pass state (8).

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.

- 2026-08-19. Advisory GHSA-h25p-f8f6-2ccf, rated moderate. The standalone HTTP mode of `@langwatch/mcp-server` 0.7.0 to 1.0.0 authorised requests by session id alone and accepted an API key in the URL. Fixed in 2.0.0 on 7 August 2026 and flagged as a breaking change. The advisory says the default stdio mode and the hosted handler were never affected, so -2 (https://github.com/langwatch/langwatch/security/advisories/GHSA-h25p-f8f6-2ccf)

## 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: the trust report and public sub-processor list on app.eu.vanta.com, whose robots.txt disallows every path
- unchecked: the PyPI project page for `langwatch`, which answered with a client challenge. The Python SDK version comes from the repository
- unchecked: dollar prices for the Growth plan. The pricing page has a currency switch and we read the euro figures only
- unchecked: whether a Restricted key shortens the MCP tool list or only refuses calls. We read the source and did not run the server
- No rate limit for the platform API and no published SLA terms were found in the reviewed documentation
- Whether the terms clause against automated or non-human access is meant to cover API and MCP use is not stated
- The DPA is available to Enterprise customers on request and was not read
- We cloned the GitHub repository before reading github.com/robots.txt. The pages later fetched from github.com are not disallowed by it

## Weaknesses

- The MCP server registers 101 tools, deletes and key creation among them, with no toolsets and no `readOnlyHint` or `destructiveHint` annotations in the source
- No request rate limit for the platform API was found in the reviewed documentation. Only two operations in the OpenAPI document declare a 429
- The status page shows Scenarios down for 9 hours 34 minutes on 8 September 2026 and trace processing degraded for 14 hours 44 minutes on 31 July
- The Terms of Service of 22 September 2026 say users will not access the platform through automated or non-human means, which the API and MCP server contradict
- The audit log is Enterprise only and records neither sign-ins nor reads of traces

## 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.

- Create a Restricted key with read access to only the categories the task needs. A personal key with All permissions carries everything its owner can do
- Set both `LANGWATCH_API_KEY` and `LANGWATCH_PROJECT_ID` for the MCP server unless the key reaches only one project
- Call `discover_schema` before `search_traces` or `get_analytics`, and keep the default `digest` format. `json` returns the full raw trace
- Allowlist MCP tools in the client. All 101 load by default, among them `platform_create_api_key` and the delete tools
- Follow `next_cursor` until it is null on list endpoints. A full page does not mean more rows exist

## 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.