mcpbeat

Add Knowledge

microsoft/add-knowledge

Add a knowledge source (public website or SharePoint) to a Copilot Studio agent. Use when the user asks to add a knowledge source, documentation URL, website, or SharePoint site for the agent to search.

8k tokens
context cost
the whole folder, loaded on every use
2
files
instructions only
0
copies elsewhere
how many repositories repackaged it
386
stars on the repo
on the repository, not the skill itself

Install

one command, takes just this skill from the repository
npx skills add https://github.com/microsoft/skills-for-copilot-studio --skill add-knowledge

What comes with it

23 601 bytes besides the instruction
knowledge-guide.md

What it tells the agent to use

found in the instruction text
Bash runs shell commands — read the instruction before connecting

The instruction itself

6 sections, as written by the author

Add Knowledge Source

Add a knowledge source to the agent. Supports Public Website, SharePoint, and Custom API sources.

Instructions

  • Auto-discover the agent directory:
   Glob: **/agent.mcs.yml

NEVER hardcode an agent name.

  • Parse the arguments to extract:
  • URL
  • Name / description (optional)
  • Determine the source type from the user's request:
  • If the URL contains sharepoint.com → use SharePointSearchSource (but first normalize the URL — see below)
  • If the user wants to connect a custom search API or database → use OnKnowledgeRequested (see Custom API section below)
  • Otherwise → use PublicSiteSearchSource
  • If SharePoint: normalize the URL before using it. Copilot Studio requires a direct folder path like:

https://contoso.sharepoint.com/sites/MySite/Shared%20Documents/MyFolder

Users often paste URLs in other formats. Handle them as follows:

| URL pattern | Example | Action |

|---|---|---|

| Direct path (/sites/.../Shared%20Documents/...) | https://x.sharepoint.com/sites/Site/Shared%20Documents/Folder | Use as-is — this is the correct format |

| AllItems.aspx with ?id= param (...AllItems.aspx?id=...) | https://x.sharepoint.com/.../AllItems.aspx?id=%2Fsites%2FSite%2FShared%20Documents%2FFolder&viewid=... | Extract and decode the id query parameter. URL-decode it to get the path (e.g. /sites/Site/Shared%20Documents/Folder), then prepend the origin (https://x.sharepoint.com) to build the direct path. Drop all query parameters (?id=, &viewid=, etc.) |

| Sharing link (/:f:/s/... or /:x:/s/...) | https://x.sharepoint.com/:f:/s/Site/EncodedToken?e=abc | Cannot convert — these are opaque sharing tokens with no extractable folder path. Ask the user: *"That's a SharePoint sharing link — I can't extract the folder path from it. Could you navigate to the folder in SharePoint, copy the URL from the browser address bar, and paste it here?"* |

Encoding rule: Spaces in the final URL must be encoded as %20 (e.g. Shared%20Documents, not Shared Documents).

  • Look up the knowledge source schema:
   node ${CLAUDE_SKILL_DIR}/../../scripts/schema-lookup.bundle.js resolve KnowledgeSourceConfiguration
  • Generate the knowledge source YAML.

Public Website:

> PublicSiteSearchSource uses Bing search to find relevant snippets within the scoped URL. It does not crawl or summarize full pages. The URL defines a search scope (not a specific page), and supports a maximum depth of 2 levels beyond the domain (e.g. https://example.com/docs/api works, https://example.com/docs/api/v2 is too deep and will be ignored).

   # Name: <Name>
   # <Description of what this knowledge source provides>
   kind: KnowledgeSourceConfiguration
   source:
     kind: PublicSiteSearchSource
     site: https://www.example.com/docs

SharePoint:

   # Name: <Name>
   # <Description of what this knowledge source provides>
   kind: KnowledgeSourceConfiguration
   source:
     kind: SharePointSearchSource
     site: https://contoso.sharepoint.com/sites/MySite/Shared%20Documents/MyFolder
  • Always include # Name: and a description comment at the top. These are important for identifying the knowledge source.
  • Save to the agent's knowledge/<descriptive-name>.knowledge.mcs.yml directory

Custom API via OnKnowledgeRequested (YAML-only)

When the user wants to connect a proprietary search API or database not natively supported by Copilot Studio, use the OnKnowledgeRequested trigger. This is a YAML-only capability with no UI designer — it intercepts knowledge search requests and lets you call a custom API to populate results.

How it works:

  • Create a topic with OnKnowledgeRequested trigger
  • Use System.SearchQuery (the AI-rewritten query) as input to your API call
  • Transform API results into the required schema: Content, ContentLocation, Title
  • Set System.SearchResults with the transformed results

Use the template: ${CLAUDE_SKILL_DIR}/../../templates/topics/custom-knowledge-source.topic.mcs.yml

Prerequisites: The user needs a connector action (in actions/) that calls their search API. If they don't have one yet, they should create it through the Copilot Studio UI or VS Code extension first.

Dynamic Knowledge URLs with Global Variables

Knowledge source URLs support {VariableName} placeholders — typically global variables — for dynamic routing based on user context. This is powerful for scenarios where different users need different knowledge sources (e.g., geolocation-specific SharePoint folders, region-specific documentation).

Pattern: Define a global variable (via /copilot-studio:add-global-variable), populate it early in the conversation (e.g., in an OnActivity/conversation-init topic based on user profile or geolocation), then reference it in the knowledge source URL with Power Fx string interpolation.

SharePoint example — a workforce agent that routes to different SharePoint folders per region:

# Name: Regional HR KB
# This knowledge source provides HR information from the {Global.UserKBURL} SharePoint.
kind: KnowledgeSourceConfiguration
source:
  kind: SharePointSearchSource
  site: =$"{Global.UserKBURL}"

Public website example — documentation site with region-specific subpaths:

# Name: Regional Docs
# Documentation for the user's region at {Global.Region}.
kind: KnowledgeSourceConfiguration
source:
  kind: PublicSiteSearchSource
  site: =$"https://docs.example.com/{Global.Region}/api"

CRITICAL: URL format for the global variable value. The variable must contain a direct, clean URL — not a SharePoint UI URL with query parameters. Copilot Studio cannot resolve AllItems.aspx links.

| Correct (set this in the global variable) | Wrong (will not work) |

|---|---|

| https://contoso.sharepoint.com/sites/MySite/Shared Documents/MyFolder | https://contoso.sharepoint.com/.../AllItems.aspx?id=%2Fsites%2F...&viewid=... |

Note: spaces are OK in the variable value (unlike static site: values which need %20), because the Power Fx $"{Global.UserKBURL}" expression handles encoding at runtime.

Implementation checklist when the user asks for dynamic knowledge:

  • Create the global variable with /copilot-studio:add-global-variable (e.g., UserKBURL, Region)
  • Ensure a topic populates it early in the conversation (e.g., via user profile lookup, geolocation, or asking the user)
  • Create the knowledge source with =$"{Global.VariableName}" syntax in the site: field
  • Remind the user: the variable must be set before any topic that triggers knowledge search, otherwise the URL will be blank

Knowledge Architecture & Best Practices

For detailed guidance on how knowledge retrieval works, source selection, content quality, security, maintenance, and advanced patterns like triggerCondition, see knowledge-guide.md.

Limitations

This skill can create Public Website and SharePoint knowledge sources.

For other types, inform the user:

> "The following knowledge source types must be created through the Copilot Studio UI as they require Power Platform configuration:

> - Dataverse tables

> - Uploaded files

> - AI Search

> - SQL Server

>

> Please create these in the portal, then clone the solution to edit them here."

How to use it

Copy the folder

Take microsoft/add-knowledge from the repository into ~/.claude/skills for personal use, or into .claude/skills inside a project.

Check the name does not clash

The agent identifies a skill by the name field in its header. Two skills with the same name cannot sit side by side — one of them will be ignored.