Daedro / Handbook
Chapters

A regional MCP network at $0

Start with an existing operator-controlled machine. Add remote sites only when the owner explicitly grants access and agrees to absorb any incremental costs. The target is zero new charges to the operator. The availability checks below confirmed no free VPS allocations. They used public price and capacity endpoints without logging into a cloud account.

What counts as a site

A site is a separately controlled physical location with an accountable owner, a known jurisdiction, a failure domain and an agreed expiry date. Several containers or VMs on one host can test concurrency and recovery. They still count as one site. A provider region that sells compute counts as a candidate location until a real resource is available to the operator.

Network size Evidence required before calling it live Safe exercise while capacity is missing
10 sites One home site plus nine other authorized locations, each reachable and verified independently. Model ten endpoints on the operator machine. Record one actual location and nine unfilled slots.
20 sites One home plus nineteen sites. Label sites that share a cloud provider, datacenter, uplink or power source. Begin with a ten-site cohort. Expand only after new operators have confirmed their resources and passed a public health probe.
30 sites One home plus twenty-nine sites. Keep a revocation contact and a teardown date for every peer. Exercise 30 synthetic endpoints locally or in granted capacity. Report the actual number of sites separately from the number of endpoints.

Home runs account services and, where approved, the managed controller on loopback. Private uploads, key material and Docker access stay under its operator's control. The billing pilot can operate in sandbox mode. Regional peers initially serve read-only public documentation or fetch the hub's public health or catalog metadata. Grant each peer that limited role. Exclude OAuth cookies, caller tokens, private package bytes, customer documents and Docker sockets from replication. A request to process customer content in another jurisdiction requires a separately approved destination and an explicitly configured service.

The current hub scales containers on one Docker host. It has no cross-site scheduler, shared encrypted ledger or automated failover. Never start two controllers against copies of the same installation state. If Frankfurt becomes unreachable, a donated Singapore mirror can still serve a public handbook. It cannot assume ownership of a private Frankfurt instance. Operations · Service trust boundaries

Ten candidate locations checked

On November 3, 2026 at 18:38 UTC, the public Vultr availability API returned the following results. The smaller SKU is vc2-1c-1gb, suitable for a synthetic network probe. The larger SKU is vc2-2c-4gb. A provider listing establishes that the SKU was offered at the time of the check. It conveys no account quota, free allocation, data-transfer permission or capacity reservation. Plan and region price catalog

Region and location 1 GiB probe available 4 GiB host available
fra · Frankfurt Yes Yes
ams · Amsterdam Yes Yes
waw · Warsaw Yes Yes
lhr · London Yes Yes
sgp · Singapore Yes Yes
nrt · Tokyo Yes Yes
syd · Sydney Yes Yes
jnb · Johannesburg Yes No
ewr · New Jersey Yes Yes
lax · Los Angeles Yes Yes

For the first cohort, choose sites where an owner has already consented. For example, an approved EU peer can test public latency from Frankfurt and an approved APAC peer can test the same URL from Singapore. The team's location alone cannot establish where its customers or their data belong. No second or third cohort has been verified.

What the equivalent rentals would cost

These figures explain why a trial or a short runtime does not establish a free network. Assume an already-funded home site, disposable probes with standard primary IPv4, 30 days, no backups for probes, and the published baseline of $5/month or $0.007/hour per probe. If the home site also needs renting, add its cost. Some locations cost more. For instance, the catalog lists $7.50/month in São Paulo. Taxes, transfer overages, retained disks, IP reservations and boot time can increase the total. Public prices · Hourly minimum and billing cap

Total sites Rented probes Continuous for a full month Eight billable hours per night One billable hour per night
10 9 $45.00 $15.12 $1.89
20 19 $95.00 $31.92 $3.99
30 29 $145.00 $48.72 $6.09

Nightly estimates multiply the listed hourly rate by hours per night, 30 nights and the number of probes. They assume each VM is destroyed after each test. A powered-off Vultr instance still incurs charges, and each creation incurs at least one billable hour. With the previous $24/month home-server scenario, continuous totals would be $69, $119 and $169 before extras. Under the current $0 requirement, every rental in this table requires an already-awarded grant that prevents cash charges or a donor taking responsibility for the bill. Stopped-instance billing · Earlier hosting comparison

Trial credits and the cash limit

DigitalOcean. The current standard offer is $5 on the first team for 90 days. A valid payment method is required. A general $200/60-day signup grant is not documented for new teams. At $6/month for a 1 GiB Basic Droplet, nine full-month probes would cost $54. A verified $5 credit would still leave $49 before extras. Overage beyond a credit is billable. This offer alone fails the $0 requirement. Signup credit · Droplet prices · Credit terms

AWS. Eligible new customers can receive $100 at signup and earn up to another $100. The Free account plan closes at the earlier of six months or credit exhaustion. Its available services and instance choices are restricted. Before counting a VM, check the plan state, each region's actual vCPU quota, instance availability, storage, transfer and public IPv4 pricing. Public IPv4 alone at $0.005/hour would consume $3.60 per address per 30-day month before any applicable allowance. The Paid plan can bill after credits run out. Joining AWS Organizations can automatically upgrade the plan and cancel remaining credits. Use only an eligible operator's genuine account and a verified provider-enforced $0 boundary. No AWS capacity or eligibility has been confirmed here. Free plan rules · EC2 eligibility · Public IPv4 prices

A credit expiry is relative to the actual activation timestamp. As an example, a November 4, 2026 DigitalOcean activation plus 90 days reaches February 2, 2027. An AWS Free plan activated November 4 reaches its six-month limit on May 4, 2027 or closes earlier on exhaustion. Copy the provider's actual UTC expiry and balance into the inventory. Use one legitimate account per provider. Never multiply grants by registering additional identities.

Admission and termination

  1. Record the owner, location, allowed traffic, budget owner, grant balance, hard-stop mechanism, maximum runtime and UTC expiry. Admit a site only with written permission to operate there. Set it aside if an attached payment method could be charged and the operator has no hard zero-cash protection.
  2. Use one nonprivileged public probe before scheduling a cohort. Verify TLS, the public hostname and a synthetic HTTP result. Track errors and round-trip latency without copying a user's request or identity to the peer.
  3. Set an independent teardown deadline before the provider cutoff, for example 48 hours early. Monitor both time and remaining credits. On a missing balance reading, revocation, unexpected resource, or upcoming expiry, stop scheduling new runs. A billing alert cannot enforce this rule.
  4. For an expiring grant, verify the remaining test resource IDs against the recorded inventory. The authorized owner removes disposable probes, disks, snapshots and reserved IPs and checks every enabled region for residual resources. Confirm teardown from the provider control plane. For a donated host, revoke the peer credential and remove its public endpoint from routing.
  5. Scale the verified cohort only after confirming the earlier cohort is within its permissions and cost boundary. A simulated plan, an SSH-accessible address and a running paid instance are distinct inventory states.

The next action at $0 is to inventory existing approved hosts and run one synthetic public read from each consenting location. Renting nodes or activating a trial is a separate decision that requires an owner, explicit terms and a verifiable no-charge stop. Hosting and private ingress