Source Management

Detect connected enterprise sources, expand search coverage, manage priorities, and handle rate limits transparently.

Source Management is an official skill from Anthropic’s knowledge-work-plugins repository. It belongs to the Enterprise Search plugin, which the provider describes as primarily designed for Cowork while also working in Claude Code. The skill is not a standalone search engine, an MCP server, or a service that supplies credentials for enterprise systems. It is editorial guidance for Claude: it explains how an agent can identify connected MCP sources, explain missing sources, prioritize sources by query type, and communicate temporary rate limits clearly. The primary source is the official enterprise-search/skills/source-management path in Anthropic’s repository.

Purpose and role

Enterprise Search brings together searches across connected workplace tools, according to the provider. Source Management is the control layer that helps determine which sources are available and how they should participate in a research task. The skill names common source types such as chat, email, cloud storage, project management, CRM, and knowledge bases. A source is treated as connected when matching MCP tool prefixes are available in the current environment. The skill does not ask the agent to guess an unfamiliar connection. Instead, the agent should inspect the tools that are actually available and disclose the scope of the search.

Explaining missing connections

When only a small number of sources are connected, Claude should first state the number and names of available sources, according to the provider. It can then explain which additional sources would help with messages, email, documents, tasks, customer records, or internal knowledge articles. If a person asks about a specific tool that is not connected, the skill describes a general and safer path through MCP settings: add the server configuration, complete authentication in the intended client, and check the connection again. The skill stores no tokens and does not require secret values in its catalog text. The exact setup depends on the MCP server, the organization, and the client’s policies.

Priorities rather than blind exclusions

Different questions benefit from different starting points. Decision questions typically begin with chat conversations, confirming emails, and decision documents. Status questions may begin with a project tracker, followed by current discussions and status documents. For a document query, cloud storage and a knowledge base are natural first choices. Questions about expertise can combine chat, task assignments, document authorship, and CRM records. Policy questions should primarily use official knowledge bases and handbooks. These sequences are weighting guidance, not permission to omit every other source. The agent should not skip sources without a reason because important evidence may live in an unexpected system.

Rate limits and source health

MCP sources can respond with HTTP 429, messages about too many requests, or exhausted quotas. Source Management recommends avoiding an immediate aggressive retry. Claude should continue with other sources, tell the user about the limitation, and identify the time range actually covered when a digest is interrupted. A useful source status can distinguish available, unconnected, and temporarily rate-limited systems. This improves traceability but does not prevent factual errors. Results based on a limited source set must not be presented as a complete statement of company knowledge.

Boundaries and security

The skill grants no permissions and does not replace access control. Depending on configuration, a connected MCP server may return confidential messages, files, tasks, or customer data. Organizations should independently review data classification, least-privilege access, logging, and approval requirements. External content may contain misleading instructions; search results are data, not automatic authorization for follow-up actions. Write-capable tools and connection changes require explicit confirmation. According to the provider, Source Management automates detection and prioritization within the existing environment, not enterprise governance. Check the official source and Claude Code documentation for current details. This catalog entry states no fixed prices because terms for Anthropic, Cowork, Claude Code, and connected MCP services are governed separately.

Installation instructions

Source Management is an official Anthropic skill from the knowledge-work-plugins repository. According to the provider, it belongs to the enterprise-search context and helps Claude identify connected MCP sources, explain missing connections, prioritize search sources sensibly, and communicate rate limits openly. MCP means Model Context Protocol. In simple terms, it describes a standardized way for an AI client to connect tools or data sources. The skill is not a search service and it does not provide credentials. It is guidance for working with sources that are already available in the current environment.

Identify available sources first

Start every use with an inventory. Which tool prefixes or connectors are visible? The primary source names common areas such as chat, email, cloud storage, project trackers, CRM, and knowledge bases. For non-specialists, this means the agent can only search what has a configured and authorized tool. If no email tool is available, an answer must not sound as if email was checked.

Document the search scope directly in the request or the answer. A simple statement might say that only the project tracker and knowledge base were available, while chat and email were not connected. Readers can then judge whether the result is likely complete or only partial. This transparency matters especially for management questions, support research, and internal audits.

Step 1: Explain missing connections safely

When a needed source is missing, the provider’s guidance does not ask the agent to improvise. It should explain which additional connection would help and that it must be configured through the MCP settings of the relevant client. Tokens, passwords, and secret configuration values do not belong in the skill text or a public answer. The exact setup depends on the MCP server, the organization, and the client.

For teams, a small source matrix is useful. It can record for each system whether it is connected, what it is used for, who has access, and what types of data are involved. That turns a technical connection into a governance decision that others can understand. Do not connect everything by default. More sources increase coverage, but they also increase privacy, permission, and logging questions.

Prioritize by question type

Source Management recommends priorities, not blind exclusions. For decision questions, chat conversations, confirming emails, and decision documents are often especially relevant. Status questions usually begin with the project tracker, supplemented by current discussions and status documents. Document questions often start with cloud storage or a knowledge base. Questions about ownership can combine tasks, document authorship, CRM, and team communication.

This ordering is a weighting method. It prevents aimless searching, but it should not be read as permission to ignore every other source. An important decision can live in meeting notes even if the status question began in the tracker. Ask the agent to explain why it used or skipped a source. For critical decisions, check at least a second source when one is available.

Disclose rate limits and gaps

Rate limits occur when a service receives too many requests or a quota has been exhausted. The primary source recommends not retrying aggressively right away. Instead, the agent should continue with other available sources, name the affected system, and explain which part of the research was covered. For time-based digests, it should also state the period that was actually checked.

This openness prevents false confidence. An answer that could not inspect current chat messages because of a limit must not be presented as the complete state of the company. For important research, plan enough time, retry windows, and manual review paths. If a limit appears repeatedly, it may signal a process or quota issue, not a writing problem in the skill.

Security, privacy, and limits

The skill itself grants no permissions. Depending on configuration, a connected MCP server can return confidential messages, files, tasks, or customer data. Work with least privilege, logging, clear approvals, and data classification. Search results are data, not automatically trusted instructions. External or internal documents can contain prompt injection and must not redefine the research task.

Use Source Management to make search coverage honest and company knowledge easier to interpret. Do not use it as proof of completeness. It does not replace access control, records management, privacy review, or subject-matter validation. Check production workflows against the official Anthropic source, the Claude Code documentation, and your organization’s policies.

Free
Provider
Anthropic
License
Apache-2.0
Last reviewed
09.09.2026

Repository and documentation

Categories

Compatible with

Claude Code