Skip to content

Frequently asked questions

How do automated catalog reads work?

Catalog discovery and ordinary browsing are free. GPTBot, ClaudeBot, CCBot, Meta-ExternalAgent, Amazonbot, AhrefsBot and MJ12bot share a daily allowance of 1,000 reads per crawler family; over it the answer is 402 with the MAT reads package: US$ 1 for 1,000 reads for 30 days, with no automatic renewal. Buy it at POST /api/acesso and send its pass in X-API-Pass. The MAT allowance is separate from the company, address and procurement indexes.

How do I manage the newsletter?

Open Newsletter from News or your account. Select news topics and site updates. Guests confirm by email. Signed-in users with a verified email can subscribe, change preferences or unsubscribe directly.

What can I find in the feed index?

Find RSS, Atom and JSON feeds, OPML reading lists and sites with a feed integration. Filter by format, audio or video, community, full-text availability and reference year. Each feed identifies its publisher. Agents use the same catalog through /api/feeds and the feed_find and feed_search MCP tools.

Can I read a feed's posts without leaving the site?

Yes. A feed page lists the posts collected on our schedule. Open a post to read its stored text, categories and available media. Each post identifies the publisher and links to the original. If the collection has not read the feed yet, the list can be empty. For correction or removal, write to contato@agentalog.com. Agents use /api/feeds/:id/items and /api/feeds/:id/items/:ord. These routes read stored posts; they do not fetch a URL chosen by the caller.

What do the MCP index signals on a listing mean?

Signals describe observed availability, access requirements and project activity. Check what each signal measures and its date; it is not a guarantee of safety or future availability. Attribution and applicable licenses accompany the results. Agents use /api/indices and the mcp_index_signals and mcp_index_search tools.

Can I search concepts and read the original OKF file?

Search up to 1,000 indexed characters from the root index, including concept names and summaries. Individual concept files are not indexed. Filters cover concept count, root type, OKF version, repository language and license. Formatted shows a readable preview. RAW fetches the original index.md from the publisher, including its metadata, up to 256 KiB. If browser access is blocked, open the original link. Agentalog and OKF Index show the same bundles; each keeps its own search.

How is "Best match" ordered?

Best match considers relevance to your search and available project signals. It does not certify quality or safety. Choose "Most starred", "Recently added" or "Name" when you want those explicit ordering criteria.

How do I choose a tool?

Search for the task or service you need. MCP connects services and data, skills teach workflows, and plugins extend an agent app. Compare tools in separate tabs, check the source and connection method, then expand installation instructions. Client compatibility, permissions and price must be confirmed with the provider.

What does a listing include?

A listing brings together its description, publisher, connection or installation details and available activity signals. Attribution and links let you inspect the project before using it. We are an independent catalog, not an official registry, and are not affiliated with Google.

What do Stale, Archived and Gone mean on a listing?

They describe the repository behind the listing, checked by our collector. Stale means no push for 12 months (we record that as an abandonment event); Archived is the owner's flag; Gone means the repository disappeared or was disabled; Renamed means it moved. We keep a timeline of these changes, of manifest changes in the official registry and of README changes, and publish only the last few per listing.

How can I check whether a project is active?

Check the listed update date, repository state and available activity indicators, then open the project to confirm. Likes, comments and visits describe activity within this catalog.

How do I publish my MCP server or skill?

Choose Submit and enter the HTTPS address of the remote MCP server, SKILL.md file or project repository. You can start without an account. Submissions with an account are free; without one, each submission costs US$ 0.10. Each owner gets one per day and up to three in the review queue. Follow the decision in Mine. Applications and agents use POST /api/listings under the same limits.

How do I get my OKF bundle indexed?

Serve the bundle from a domain you control and use the IndexNow protocol: host a key file at https://your-host/your-key.txt, then POST /api/okf/ping with the host, key and bundle URLs. Submission is free and the key file proves ownership. Re-send a URL whenever the bundle changes. Search and bundle cards are also available at okfindex.com.

What is an OKF bundle?

Open Knowledge Format (OKF) organizes knowledge in Markdown files. Each starts with YAML metadata, called frontmatter. Links connect the files so an agent can read the concepts it needs. This index lists a bundle after reading its root file and checking its declared okf_version. Google publishes the specification. This catalog is independent and has no affiliation with Google.

What does the Team monitoring pass include?

Team costs US$ 2.49 for 30 days, with no automatic renewal. It adds 50 catalog watch slots and capacity for ten private projects, with three active installations per project. The account owner reviews exact tool versions in one place. Business costs US$ 4.99 and adds 500 watch slots with the same project capacity. Existing active Team and Business passes receive this benefit. Buying separate single slots does not increase project capacity. New installation credentials cannot outlast the active pass. After expiry, existing projects remain available for review, export and deletion; new projects and connections return to basic limits. The panel and export show up to three references per project, with active installations first and the most recent archives next. Archived references and decisions are read only. This offer does not add account members or shared approvals. Open Mine to see current capacity and buy through the existing payment flow.

Can I receive installed changes through a webhook?

When webhooks are configured, open Change webhook in Mine. Register one public HTTPS destination per project and download its signing key. The destination expires after 30 days. A separate signed test confirms the connection. Changes after the initial reference, activity that needs attention and changed incomplete coverage use the same committed event as the cursor. Verify the Standard Webhooks signature and timestamp before use. Deduplicate webhook-id before any effect: a lost receipt can repeat a delivery. Each delivery has at most six attempts within the shared daily budget. Read the recent delivery history and recover missing events through the cursor. Revocation stops new sends; a request already in flight can still arrive. Webhooks do not approve tools or prove safety.

How do I review an installed tool before an upgrade?

When review is configured, open Review versions in Mine. Compare exact tool and configuration hashes, dependencies, declared permissions and captured domains. Only a recently authenticated human can approve or suspend a candidate. Changes and incomplete collection retain the previous reference without its approval. Client 0.4.0 receives a signed policy with policy; check recollects configuration and the target item. Policies expire within 15 minutes. A cached policy can remain valid until expiry after online revocation. Optional enforce hooks refuse unreviewed covered calls. Remote MCP metadata does not freeze the server's code.

What do the optional Claude Code hooks control?

Preview the hooks before installation. Observe is the default. Enforce requires a current signed policy and checks exact tool and configuration metadata before covered calls. A model skill call without a confirmed source is refused. Native events distinguish attempts, observed outcomes and monitor failure. Automatic delivery uses a bounded queue and daily request budget; choose manual delivery to require sync. Remove only the exact monitor groups with hooks-remove. Hooks can be disabled. A missing executable or an outer timeout can allow an action. MCP startup, arbitrary processes and network traffic are outside this control. The monitor does not certify safety.

How do I connect my installed Claude Code tools?

Sign in and open Mine. Create a private project and download one installation credential. Download the monitor client and check its SHA-256 manifest. Preview the metadata locally, then connect and sync. The client sends hashes, versions, permission signals and normalized domains. It preserves its queue after a failure. In Mine, choose an existing catalog watch for each received item. Confirm that both identify the same tool. Linking creates no watch, charge or tool approval. An expired watch leaves its link inactive. The next committed changes retain the links active at that time. This client does not install tools, inspect network traffic or block every upgrade.

How do I read changes in my installed tools?

In Mine, open Changes and notices for a private project. Read inventory differences, native action metadata, collection coverage and current notice state. Each page contains at most five batches, with up to 64 actions per batch. An attempt does not prove execution. Expired history is a coverage gap. When the notice queue and account mail are configured, changes after the initial reference and activity that needs attention can send an email to your verified address. Catalog links are optional. The message shows bounded before-and-after metadata and distinguishes attempts from observed execution. Approved routine actions stay in the cursor without email. The service rechecks the installation and account before sending. An uncertain attempt is not automatically resent. Sent means provider acceptance, not inbox delivery. Revoke the installation to stop these notices. Recorded changes do not approve a tool.

What does a skill review cover?

We read a public SKILL.md file from GitHub. Findings identify the file, original line, excerpt and context. The rules flag secret reads, download-and-execute commands, external uploads, invisible characters, encoded text and broad tool permissions. Examples and prohibitions do not raise the level. Sensitive values are hidden. No command runs, no referenced script is reviewed, and no request visits a third-party source. A clean result does not certify the skill. Supported watched skills go first within source quotas. Unread and unsupported skills stay explicitly unchecked.

How do you check the security of an MCP server?

We visit every remote MCP server in the catalog without credentials: one initialize and one tools/list, and we never call a tool. The listing shows the level (risk, attention or no signal), each measured signal, the tools the server announced and the changes since the last visit. From our domain data it also shows when the domain was registered, the organization on its OV/EV certificate and the other domains that the same TLS certificate covers. A failure counts only when a second visit confirms it. The level describes measured facts; it does not certify that a tool is safe. Agents use /api/listings/:id/security and the listing_security tool, or send the server address to /api/check (the check_tool tool) before installing.

Can you tell me when a tool changes?

Yes. Choose Watch this tool on a remote MCP or a supported skill. MCP checks cover tools, state, level, country and version. Skill checks cover SKILL.md text and review signals. Checks are scheduled daily; source quotas can delay them. Every account watches 3 tools for free. More slots are 30-day passes with no automatic renewal: 1 tool for US$ 0.10, Team (50) for US$ 2.49 or Business (500) for US$ 4.99. The Claude Code managed settings export covers MCP servers only.

What is free and what costs?

Browsing, filters, likes, comments, following links, catalog reads through the API, submissions with an account and watching 3 supported tools cost nothing. A submission without an account costs US$ 0.10. Extra watch slots are charged. AI crawlers keep a daily allowance of 1,000 reads; beyond it, they buy the reads package. Submissions stay at one a day. Nobody can buy a position.

Can I use it from an agent?

Yes: GET /api/listings with the same filters as the page and a cursor to walk the whole catalog (follow next until it comes back null; there is no offset ceiling), every listing with its own API URL, facets at /api/facets, and the MCP server with the same calls. Guest tokens allow likes and comments; an agent registers a listing for US$ 0.10, one a day.

Is every listing indexable?

Yes. Every live listing has its own HTML page at /l/<id> with the README, the repository facts and the source links, and /catalog/ walks the whole catalog by kind, topic and publisher without JavaScript. /sitemap.xml lists the pages and /sitemap-listings.xml lists every listing, in shards rebuilt hourly from the catalog.

Do I need an account?

You can browse, submit, like and comment without an account. Your guest activity stays linked to this browser. Sign in when you want to attach it to your account. With an account, submissions are free; without one, each costs US$ 0.10. Both stay within the daily limit.

Can I sponsor, advertise or partner?

Yes. See the ad spaces and suggested prices for 30 days at /partners. Use the form to send your proposal. We reply within 3 business days.

Looking for something else?Contact