Celestia Datacenters in Pulse, provider incidents, and the same data for AI agents through BlobPedia

Hi everyone! This post covers three things we built for operators: Celestia Datacenters in Pulse, a catalogue of rentable servers checked against Celestia’s requirements; provider incidents; and access to the same data for AI agents over MCP, through BlobPedia, our Celestia knowledge base.

1. Celestia Datacenters in Pulse

Why. Choosing a server for Celestia means going through different providers, checking the CPU against the v6 list, sizing disk and traffic for a validator, bridge or Fibre, and checking whether the provider is excluded from FDP and whether the server is outside the EU/NA. We brought all of this into one place: Celestia Network Dashboard - Validators, DA Layer & Bridge Nodes | Pulse by Unity Nodes

Catalogue. 1,427 servers from 23 providers, 33 countries and 6 regions. Most of the data comes from the providers’ official APIs, the rest we collect ourselves. The indicator at the top shows how many providers answered on the last refresh. You can switch between Mainnet Beta and Mocha.

A verdict for every server and role. For each server and each role (validator, bridge, Fibre) Pulse gives a verdict: Compatible, Needs benchmark, Check caveats or Not compatible, always with the reason. In the screenshot, for example: the CPU passed Celestia’s v6 survey at 1.7 times the required speed.

Three role modes

  • Consensus / Validator and Bridge, each with two presets: Official docs (taken verbatim from the documentation) and Recommended minimum (below the docs, but not below what runs reliably).
  • Fibre: requirements are derived from your stake. Find your validator and the calculator shows assigned rows, storage, ingress, memory and the recommended link, and servers in the catalogue are checked against these numbers. A “November preview” setting multiplies the load by 15. That multiplier is our own estimate, not an official Celestia figure.

CPU. Models from the tested list in the v6 release notes are marked separately as recommended. For everything else we point to tools/cpu_requirements, since core count alone does not predict the result.

Server card

We collected nearly everything providers publish about their servers:

  • price, setup fee and deploy time
  • upgrade options with prices: disk, memory, bandwidth
  • traffic: quota or unmetered, and how inbound traffic is billed
  • whether the provider fits FDP: Hetzner and OVH are excluded, EU and North America are discouraged
  • how many Celestia nodes already run at that provider, for a general sense of decentralisation
  • open provider incidents (section 2)

Tools

  1. Filters by region, country, provider, price, RAM, storage, CPU, port and traffic.
  2. Sorting and comparison: sort by best fit or price, pick several servers and compare them side by side.
  3. Shareable filter links: set the parameters you need and share the link; the page opens for your colleagues with those parameters already applied.
  4. Toggles:
    • Hide auction hides auction offers: discounted servers that can disappear at any moment.
    • Include out-of-stock also shows servers that are currently unavailable, so you can see a provider’s full range.
    • Hide FDP-excluded hides providers excluded from the Foundation Delegation Program.
    • Compatible only keeps only servers that meet the requirements of the selected role and preset.

Alerts for new servers. “Alert me on new matches” connects your current filter to a Telegram bot. Set RAM, storage, traffic, toggles, a region, a country or specific providers, and the bot tells you when a matching server appears.

2. Provider incidents

A cheap server is a bad deal in the middle of a provider outage. Pulse reads the public status pages of some hosting providers, for example OVH, Hetzner and Cherry Servers, keeps the incidents it judges relevant to servers and network, and shows the open ones in the server card. Separately, for over 9 months we have been collecting public information on mass provider outages and security events, with sources.

Two limits: Pulse reads the status pages of only some providers, so “no open incident” is not proof that there is none, and these feeds rarely mark old items as resolved, so read the dates.

3. The same data over MCP

Everything in sections 1 and 2 is also available to AI agents. MCP is the protocol that lets an assistant such as Claude, Cursor or Windsurf call tools. BlobPedia’s MCP server relays Pulse’s data, so an agent can do in one conversation what an operator would otherwise do by hand. Connect with one command, no key needed:

claude mcp add --transport http --scope user blobpedia https://blobpedia.org/mcp

Setup for Claude Desktop, Cursor, Windsurf and VS Code is at BLOBPEDIA for AI | MCP + REST API | BLOBPEDIA.

The agent does not judge compatibility. Pulse does, through an endpoint that runs the same code as the Datacenters page, so the page and the MCP answer cannot disagree. One server’s detail includes the provider’s open incidents:

And size_fibre_node returns the extra hardware Fibre needs for a given share of voting power:

Fibre readiness from Cumulo. With Cumulo’s agreement, BlobPedia also relays data from their Fibre Network Monitor: readiness against the 2/3 threshold from four probe regions, quorum latency, data retention checks, blobs and fees on Mocha. As we said when it was announced, this is excellent work, and thanks to it our MCP now carries more important information for all of us.

Prices, stock and incidents are a snapshot refreshed every 15 minutes, and each answer names its source and, except the Fibre sizing calculator, its time.

4. BlobPedia

The tools above sit on top of a knowledge base that existed before. BlobPedia indexes 5,540 records from 12 sources: CIPs with their status history, documentation, forum threads, GitHub issues, pull requests and releases, video transcripts, blog posts, whitepapers and ecosystem projects. An agent can search it by keywords or by meaning, open a CIP, a forum thread or a doc page, and check live chain state. In total there are 63 MCP tools, a REST API and the website https://blobpedia.org, with no key needed.

What this does not do

  • Prices and stock come from providers’ public catalogues and change. An answer is a starting point, not a quote: check the terms on the provider’s site before ordering.
  • Verdicts follow Pulse’s rules as they are today, and those rules change as Celestia’s requirements change.
  • Provider incidents come only from the status pages Pulse reads.
  • Fibre figures cover the Mocha testnet only.

What’s next

We actively maintain our products after launch and will keep developing them. Soon we plan to add a Hardware tab to Pulse that shows the real load of a validator, a bridge and a Fibre node, to complement the information a validator needs when launching.

Tell us what is missing

We would be grateful for your feedback: are the role thresholds right, which providers or features are missing, where do you see inaccurate data, and which operator question the tools cannot answer yet. Please reply in this thread.

2 Likes

Update: what’s new in Celestia Datacenters since launch

A month after launch, here is what has changed in Datacenters:

More providers. The catalogue now pulls from 36 providers instead of 23, with over 1,600 servers in 37 countries. New ones include 23M, AlbaHost, Allnodes, ColoCrossing, GameServerKings, IP-Projects, IPAX, KoDDoS, ServeTheWorld, ServerMania, SwissMade.Host and Tier.Net. The “Live” indicator shows how many providers answered on the last refresh.

Several roles on one server.

You can now tick more than one role, for example a validator and Fibre on the same server, and the requirements are summed. Each preset shows its numbers: “Official docs” follows Celestia’s hardware page, and “Recommended minimum” shows the date it was measured.

Map.

Offers by city. Click a city to filter by its country.

Server cards.

Each card explains its verdict line by line: CPU against the v6 survey (with a link to the community HCL), RAM, disk, traffic terms, the provider’s FDP status (now four levels: eligible, discouraged, not advised, excluded), whether Celestia nodes already run at that provider, and stock.

Servers that do not fit a role are now listed too, after the compatible ones, with a red verdict and the reason, for example a CPU that failed the v6 benchmark or too little disk. The “Compatible only” toggle hides them.

Provider reliability.

Each provider has a popover with the outages and security events we have on record and how many validators we observe there. No record is not a guarantee, and the page says so.

Compare.

Pick a few servers and compare them side by side.

Fibre sizing.

Fibre requirements follow your stake on the selected network: assigned rows, storage, ingress, monthly traffic and worst-case RAM, from the Foundation’s Fibre calculator.

Real usage next to the requirements. We have also launched a Hardware tab that shows what our own nodes use per role, next to these requirements: Pulse Hardware: measured resource use of Celestia nodes, per role

For AI agents.

The same data is available through the Blobpedia MCP ( BLOBPEDIA for AI | MCP + REST API | BLOBPEDIA ): list_datacenter_offers, get_datacenter_offer, list_datacenter_providers and size_fibre_node.

Alerts for new matching servers work as before.

What would you like to see next: another provider, a filter, a role?