> ## Documentation index
> Fetch the complete documentation index at: https://context7.com/docs/llms.txt
> Ask it with GET https://context7.com/docs/ask?q=<keywords or question>. Short keywords rank best.
> Use these to discover all available pages before exploring further.
> Feature requests: https://context7.com/docs/resources/feature-requests.md lists what users asked for and how to vote for a request or add one.
> Page rating: `GET https://context7.com/docs/feedback?path=/docs/enterprise/authority-labels&helpful={true|false}&comment={comment}` records whether this page helped. The `comment` (URL-encoded, at most 500 characters) says what helped, or what was missing or wrong; a `helpful=false` rating needs it. It is anonymous and needs no key. Comments hold no code, secrets or personal data.

# Authority Labels

> Label library content as a standard, a guideline, or an example, so AI agents can tell authoritative content from informal content. Exclude content that must not appear in results.

AI agents that read your documentation cannot tell an approved standard from an old draft or an informal note. Authority labels let your team say which is which. You define a short list of classes, such as `standard`, `guideline`, and `example`, and you write a definition for each one. Then you assign classes to libraries, or to paths inside a library. Each result with a class shows that class, and the response defines each class that appears.

Only admins set classes. No model assigns or changes a class, and the reranking model never sees classes. Agents see the class names and definitions in results. Authority never grants access to a library.

## Enable authority labels

Authority labels are off by default. Open **Workspace Settings**, select **Authority labels** under **Access**, and turn on **Enable authority labels**. Labels, ordering, and exclusions then apply to every query through the MCP server, the API, and the dashboard snippet search.

You can prepare classes and rules while the feature is off, in the dashboard or in the [GitOps manifest](/enterprise/gitops). Rules that are saved while the feature is off do not change any result.

When you turn the feature off, Context7 keeps your classes and rules. Content that your exclusion rules hide becomes eligible for retrieval again, and library access control still decides who can query each library. If exclusion rules exist, the dashboard asks you to confirm first. The audit log records each change to this setting.

<Frame>
  <img src="/images/enterprise/authority-labels/settings.png" alt="Authority labels settings with the Enable authority labels switch off and the class list" />
</Frame>

## Define the classes

Open **Workspace Settings** and select **Authority labels** under **Access**. The classes are on the same card as the **Enable authority labels** switch. While the feature is off, the card marks it as off, and you can still edit the classes. A new instance starts with three classes:

| Class | Default definition |
| --- | --- |
| `standard` | An approved standard or requirement. Follow it unless a newer standard replaces it. |
| `guideline` | Recommended practice. Follow it by default and deviate only with a clear reason. |
| `example` | An illustration of how something was done. It is not a requirement. |

You can rename a class, change its definition, add a class, and remove a class. The order of the list is the order of authority: the top class is the most authoritative. To change the order, drag a class by its handle. With a keyboard, focus the handle and press the up or down arrow key.

You cannot remove a class that a library rule still uses. Change or delete those rules first.

## Assign classes to a library

Open the library, click **Settings**, and find the **Authority** section. When authority labels are off for the workspace, the section shows a notice: you can still edit and save rules, but they do not apply.

- **Library default**: the class for all content in the library, or **Excluded** to hide the whole library.
- **Path rules**: a class for part of the library, or an exclusion. A rule is a **Prefix** or a **Glob**.

Click **Preview** to see which pages a rule matches before you add it. A saved rule that matches no indexed page shows a **no matches** badge. To add a path rule, click **Add**, then click **Save changes**. Context7 saves the whole set of rules together in one request, so a new or changed rule applies only after you click **Save changes**.

<Frame>
  <img src="/images/enterprise/authority-labels/library-rules.png" alt="Authority section of the library settings with a library default, path rules, a no matches badge, and a rule preview" />
</Frame>

### What a rule matches

A pattern that starts with `https://` or `http://` matches the page URL, the same link that appears on the `Source:` line of a result. For Confluence, this is the link that opens the page in your browser. Any other pattern matches the page path:

| Source | Page path |
| --- | --- |
| Git repository | The path inside the repository, for example `/docs/standards/release.md` |
| Website, Confluence, llms.txt, OpenAPI | The path of the page URL, for example `/wiki/spaces/QA/pages/123` |
| Google Drive | The Drive folder path and file name, for example `/QA/Standards/Release sign-off`. A slide uses the path of its presentation. |
| Notion | The title path of the page inside the selected pages and databases, for example `/QA Handbook/Standards/Release sign-off`. A `/` in a page title becomes `／`. |

The path of a page URL includes its query string, if it has one.

A path pattern without a leading `/` gets one, so `standards/` and `/standards/` are the same rule.

A prefix matches whole path segments. `/drafts` matches `/drafts` and `/drafts/plan`, but not `/drafts-old/plan`. In a URL pattern, a `?` or `#` after the prefix also ends the segment. To match the start of a file or folder name, use a glob, for example `/drafts*/**`.

<Warning>
Google Drive and Notion paths come from folder names and page titles. When someone renames or moves a page or folder, its path changes, and a rule can stop matching it. This applies to exclusions too. Check for the **no matches** badge after you reorganize content, and prefer a URL rule when a page must keep its label.
</Warning>

In a glob, `*` matches any characters inside one folder, `**` matches any characters across folders, and `?` matches one character. A glob must match the whole path. For example, `/qa/**/checklist-*.md` matches `/qa/release/checklist-launch.md`.

### Which rule wins

When more than one rule matches a page, Context7 applies them in this order:

1. **An exclusion always wins.** If any exclusion matches the page, including a library exclusion, the page is excluded.
2. **The most specific class rule wins.** Specificity is the number of literal characters in the pattern. A longer prefix is more specific than a shorter one. For a glob, `*` and `?` do not count. Specificity counts the whole pattern, so a URL pattern, which includes the host, usually beats a path pattern. If two rules have the same specificity, a prefix wins over a glob, and then the older rule wins.
3. **The library default applies** when no path rule matches.
4. **The page has no class** when no rule matches and the library has no default.

For example, with these rules on one library:

| Rule | Class |
| --- | --- |
| Library default | `guideline` |
| Prefix `/standards/` | `standard` |
| Prefix `/standards/drafts/` | Excluded |

`/standards/release.md` is a `standard`, `/guides/testing.md` is a `guideline`, and `/standards/drafts/new.md` never appears in results.

## Exclude content

An excluded page or library never appears in results, through the MCP server or the API. A library excluded as a whole also disappears from library search, and a documentation request for it returns the same "not found" response as a library that is not indexed.

Exclusions apply to content that is already indexed. You do not need to re-index or delete the library.

## Changes apply at once

Context7 resolves classes when it answers a query, not when it indexes content. A change to a class, a definition, or a rule applies to the next query. You do not need to refresh the library.

## How authority changes the order of results

Authority only reorders results that are already relevant to the query. It never adds a result, and it never moves a weakly related result above a clearly more relevant one.

- **With reranking on**, Context7 reorders only the results that the reranking language model selected as relevant. The other results stay below them.
- **With a decision model for reranking**, the model ranks every result, so authority does not change that order. Results still show their labels, and exclusions still apply.
- **With reranking off**, a result can move ahead of another only when their relevance scores differ by at most 0.02. The same rule applies when reranking fails or has too few candidates.

The 0.02 band is fixed. It was measured on `text-embedding-3-small`, where it lets authority swap near-ties but does not move a result past a clear drop in relevance. Other embedding models spread their scores differently, so the same band can swap fewer or more results with them.

Within those limits, a more authoritative class comes first, and content without a class comes after all classes. Results with the same class keep their original order. When no rules exist, the order does not change.

## What agents receive

The response from `query-docs` and from [`GET /v2/context`](/enterprise/api/context/get-documentation-context) starts with a legend that defines each class that appears in the results. Each labeled result has an `Authority:` line, and each result that has a source URL has a `Source:` line with that link.

```text
Authority classes in these results:
- standard: A mandatory QA requirement. Every release must follow it.
- guideline: Recommended practice. Follow it by default and deviate only with a clear reason.

--------------------------------

### Release sign-off checklist

Authority: standard
Source: https://github.com/acme/qa-standards/blob/main/standards/release-signoff.md

Every release candidate needs a passing smoke suite and a sign-off from QA.

--------------------------------

### Guidelines > Exploratory session

Authority: guideline
Source: https://github.com/acme/qa-standards/blob/main/guidelines/exploratory.md

Run a 30-minute exploratory session for each new feature.
```

A result without a class has no `Authority:` line. The legend lists only the classes that appear, so definitions are not repeated for each result.

The snippet search in the dashboard (`GET /api/projects/{library}/search`) returns the same information as JSON. Each result in `codeSnippets` and `infoSnippets` has a `sourceUrl` and an `authority` field. The example shows one entry of `codeSnippets`, with some fields left out. It is not a full response:

```json
{
  "id": "signoff",
  "score": 0.8642,
  "title": "Release sign-off checklist",
  "pageId": "https://github.com/acme/qa-standards/blob/main/standards/release-signoff.md",
  "sourceUrl": "https://github.com/acme/qa-standards/blob/main/standards/release-signoff.md",
  "authority": {
    "class": "standard",
    "definition": "A mandatory QA requirement. Every release must follow it.",
    "rank": 1
  }
}
```

`authority` is `null` for a result without a class. For an excluded library, the response has `excluded: true` and empty `codeSnippets` and `infoSnippets` lists.

## Audit log

Every change to a class, the class order, a rule, or the **Enable authority labels** setting is written to the audit log with the admin who made it. Open **Workspace Settings**, select **Audit log** under **Overview**, and filter by **Authority class**, **Class removed**, **Class order**, **Authority rules**, or **Authority labels**.

## Who can change labels

Only admins can view and change classes and rules. Everyone who can query a library sees its labels in results, including anonymous MCP and API callers when MCP authentication is not required. [Library access control](/enterprise/security/library-access) still decides who can query each library.
