Wiki Intelligence for Agents

Structured Fandom and MediaWiki search, retrieval, revision monitoring, and bounded knowledge exports for agents and automation.

Wiki Intelligence calls a public MediaWiki Action API over HTTP. It returns compact JSON records with page IDs, canonical source URLs, revision timestamps, and explicit bounds. It supports Fandom-hosted wikis and compatible public MediaWiki installations; API extensions and site configuration vary.

What it does

• Search pages and retrieve normalized page text, categories, redirects, and revision provenance.
• Extract common Infobox template parameters with ambiguity and malformed-template flags.
• Enumerate category members, backlinks, and recent changes.
• Inspect revision metadata and compare two revisions with a bounded diff.
• Crawl a category to a caller limit (maximum 500 pages per run, recursion depth maximum 3).
• Export only category pages changed since a required timestamp.

One dataset record is written per useful hit/page/change. Records include operation, source, retrievedAt, warnings, and the operation-specific data object. Content is plain text where practical; infobox parsing is best-effort because templates are community-specific.

Example inputs

Search:

{
  "wikiUrl": "https://minecraft.fandom.com",
  "operation": "search_pages",
  "query": "redstone comparator",
  "limit": 10
}

Incremental category export:

{
  "wikiUrl": "https://terraria.fandom.com",
  "operation": "knowledge_export",
  "category": "Bosses",
  "since": "2026-09-20T00:00:00Z",
  "limit": 100
}

See examples/ for more operations and representative output.

Limits and compatibility

• limit is 1–200 for ordinary operations and at most 500 for crawl/export.
• Category recursion is limited to depth 3; loops and duplicate page IDs are suppressed.
• API calls are sequential, rate-spaced, retried for transient errors, and time-limited.
• Text output is capped at 500,000 characters per page; response bodies are capped at 5 MB.
• knowledge_export requires since and returns pages in the selected category that have a matching recent-change record. It does not maintain durable state between runs.
• Arbitrary endpoints must be public hosts resolving only to public IP addresses. Private/local hosts and unsafe redirects are rejected. Do not treat this as a substitute for network egress controls in a self-hosted deployment.
• Private wikis, disabled APIs, custom API paths, and extensions that omit expected fields may not work. For nonstandard installs, pass the exact public api.php in apiUrl.
• Some APIs may suppress editor identities or revision content. Infobox layouts are not standardized; fields are returned with confidence/warnings rather than guessed.

Attribution and licenses

The Actor does not own wiki text. It returns source URLs, page IDs, revision IDs, and timestamps so downstream systems can retain attribution and identify the precise source version.

Fandom says most wiki text uses CC BY-SA 3.0, while some communities use other licenses, including noncommercial variants. The license shown by the individual wiki governs reuse. Before redistributing or commercially using exported text, verify that wiki's license and the source terms, give attribution and a link to the original page, identify changes where applicable, and comply with ShareAlike or NonCommercial terms. API availability does not grant extra reuse rights. Images and other media may have separate rights and are not included as media files by this Actor.

Start with Fandom's licensing policy, its reuse guidance, and the target community's own license page. For standard MediaWiki installations, check the site-specific copyright/license page.

Pricing and spending limits

The proposed PPE event is knowledge-unit, charged once when one result is written. Proposed launch price: $0.001 per record ($1 per 1,000). No charge is emitted for unsuccessful work. Keep Apify's synthetic apify-actor-start event at its default setting; configure only one result event to avoid double charging. Suggested minimum run charge limit: $0.001. Detailed assumptions and sample economics are in PRICING_NOTES.md.

Set a per-run maximum charge in Apify before running. The SDK checks the result of each output charge and stops at the platform spending boundary. Usage costs are not passed through as a separate line item in the recommended pricing model. Agentic payments additionally require account KYC, which is an account-owner step.

Run locally

npm install
npm test
npm start

For local execution, provide INPUT.json using Apify local storage conventions or run with Apify CLI. The Actor requires outbound HTTPS/DNS access to the target wiki and uses no proxy, browser, external API key, or telemetry.

Build and deploy

1. The private GitHub repository is TheHoodedMan04/wiki-intelligence-actor. It is intended to contain the contents of this folder (not the Money Goblin project parent); .gitignore excludes local dependencies/storage.
2. In Apify Console, go to Actors → Develop new → Import from Git → GitHub, select this repository and main, and leave the Actor unpublished/private. The account-owner may need to complete the GitHub authorization flow.
3. In the new Actor's Source tab, verify the repository URL and branch. For a private Git source, follow Apify's deployment-key prompt: copy the read-only public key and add it as a deploy key to this one GitHub repository.
4. Inspect the .actor configuration. Before choosing Build, check the account's platform usage and confirm the build/test will not incur paid usage. A build creates a platform image and may use platform resources; no build was started here.
5. If continuing with private QA, configure the Actor's permissions as limited and keep Standby off. Configure PPE with knowledge-unit at $0.001 per stored record and keep the synthetic start event at its default. Remove/disable synthetic dataset-item pricing if the Console shows that it would double-charge custom events. Set a $0.001 minimum max-total-charge amount and do not pass usage through.
6. Run the fixture suite locally. If platform usage is acceptable, run low-limit Wikipedia and Fandom samples with a strict max charge, inspect run cost and output, and revise PRICING_NOTES.md from measured usage.
7. To submit publicly later: open Development → My Actors → [Actor] → Publishing, complete display information, monetization, sample output, output schema, and permissions, then select Publish on Store only after human approval. Actor listing visibility and source-code visibility are separate; Apify documents that source built from a private repository cannot be made public. If you want public source files, first move to a public repository or Apify-hosted source and review the license.
8. Complete payment/KYC account steps yourself if desired; they are required for payouts and agentic-payment eligibility and are outside this project.

No public or private Apify Actor, Apify account change, or Apify run was created. The private GitHub repository has been created and populated with this repo-ready project. Apify's GitHub import flow and private-repository deployment-key steps were verified in current docs; an Apify connection/build must be completed by the account owner.

Local QA

Run npm test for deterministic mocked API and fixture tests. Public network smoke tests are reported separately in QA_REPORT.md.

Store listing draft

• Suggested title: Wiki Intelligence for Agents
• Suggested slug: wiki-intelligence
• Short description: Search, retrieve, monitor revisions, and export bounded, provenance-rich knowledge from public Fandom and MediaWiki sites.
• Tags/categories: AI, Automation, Developer tools, MCP servers, MediaWiki, Fandom, knowledge export
• Pricing: Pay per event, one knowledge-unit per output record, proposed $0.001 / unit.

Full listing text and release checks: STORE_LISTING.md, RELEASE_CHECKLIST.md.