An AI agent’s answer from the Raku Knowledge Base is easier to verify when it identifies three things: the page it found, the actual heading containing the evidence, and the passage that supports the claim. The documented MCP tools can help locate and retrieve pages, but they do not establish line numbers or passage IDs. Treat a search result as a route to evidence—not proof that the answer is correct.
What the Raku Knowledge Base MCP tools return
The official Raku Knowledge Base MCP connection guide documents three tools:
![The Microsoft Copilot Bible: [3 in 1] A Beginner's Guide to Harness AI in Office 365 to Automate Workflows, Generate Content, Analyze Data, and 10x Your Productivity](https://m.media-amazon.com/images/I/41+5Qv4pd2L._SL160_.jpg)
As an Amazon Associate I earn from qualifying purchases.
raku_kb_searchaccepts a query and limit, then returns matching pages with a URL, title, section, and snippet.raku_kb_pageaccepts a URL or path and returns the page text.raku_kb_sectionslists top-level sections and page counts for the index the server has loaded.
These results make it possible to trace an answer from search to a document and then inspect the document itself. They do not, by themselves, show that a quoted passage supports a particular claim. A page citation is only useful when the cited words actually establish the point being made.
How to verify an agent’s answer
- Search for the question or a distinctive phrase. Keep the returned page URL and snippet as leads, not as a final citation.
- Retrieve the page. Use its returned URL or path with
raku_kb_pageand inspect the page text. - Locate the real heading. Read the page structure rather than guessing a section from the page title or the search result’s section field. The indexed record is described as containing a top-level section, not a full heading path.
- Check the supporting words. Ask the agent to quote the relevant passage or identify it accurately, then confirm that it supports the specific claim. The first-party blog says returned page text has no line numbers, so a demand for line-number citations is not supported by the documented output.
- Check the corpus boundary and freshness. A failed search means the answer was not found in the loaded collection; it does not establish that Raku documentation has no answer.
For a prompt, ask the agent to answer only from the MCP corpus, provide the page URL and heading, quote the words it relied on, and say when it cannot find support in that corpus. That request is more verifiable than asking for line numbers the service does not return.
#1 Best Overall
Example: tracing a grammar answer to the right section
Suppose the question is, “How does an action class share state between grammar rules?” A first-party blog post published September 23, 2026, describes searching the collection, opening /doc/language/grammars, and finding the relevant explanation under “Always succeed” assertion—not the tempting “Attributes in grammars” section.
In the Digifier and Devanagari example, the action class has an @!numbers attribute. Its succ method pushes an empty slot, and digit appends to the last slot. The methods therefore share state through the action object’s attribute. The nearby “Attributes in grammars” discussion describes a different boundary: grammar attributes are reachable only from methods, and a change made in a method called from a token affects that token’s own match object.
This is why the page alone is not enough. The page is relevant, but the actual heading and supporting code passage make the answer checkable and prevent a plausible section guess from being mistaken for evidence. See the Raku Knowledge Base blog post for the worked example.
What the index covers—and what a miss means
The connection guide describes the corpus as official Raku documentation, module documentation from both ecosystems, and the examples repository. The material is written in Podlite, with structural elements such as sections preserved through the build. That scope is useful, but it is not a promise that every page on the live knowledge-base site or every resource about Raku is indexed.
Rank #3
A September 25, 2026 article by Alexandr Zahatski notes that the MCP collection does not automatically include every page published on the knowledge-base site or all tooling around it. When a search finds nothing, check the relevant area of the official Raku documentation site before concluding that the subject is undocumented.
The same article reports these dated measurements of the source tree:
Rank #4
| Measure | Reported value and qualification |
|---|---|
| Files in source tree | 60,501, measured by Alexandr Zahatski on September 25, 2026. |
| Candidate files in a format that could produce a page | 20,469, measured by Alexandr Zahatski on September 25, 2026. |
| Files outside those page-producing formats | 40,032, measured by Alexandr Zahatski on September 25, 2026. |
| Sample of published pages sourced from Raku programs | 219 pages; every page in that sample contained embedded RakuDoc, according to the September 25, 2026 article. |
These are measurements reported in that dated article, not live totals. Its description of a September 1, 2026 ecosystem report characterizes it as a list of modules added or updated with version changes and links—not a ranking.
The Tool Desk
Outbyte PC Repair FREEClear out junk files and repair common Windows errorsFree Scan →Outbyte Driver Updater FREEScan for outdated or missing drivers - takes under a minuteDriver Scan →How fresh is an offline index?
The offline bundle distributed with data releases includes a prebuilt search index. The official guide warns that the bundle reflects the snapshot taken when its data release was cut and may trail the live site by up to a month. Counts returned by raku_kb_sections describe the index actually loaded, rather than necessarily matching the current website.
To update the offline data, the guide directs users to download a newer release whose tag begins with data- and unpack it over the existing directory. Its installation example uses the bundled raku-kb-mcp-offline.tar.gz, Node.js, npm install, and MCP client configuration. The guide’s claude mcp add command is specific to that client; other MCP-capable clients use their own configuration while pointing to the same server file. Consult the current guide for the release and client-specific setup details.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
