Do you know you can find all the ContextVM-enabled MCP servers listed on our website?

You can also try them directly through our LLM chat!

Check them out!

https://contextvm.org/servers

Building in FOSS is beautiful and powerful. Everything you do can impact upstream dependencies and downstream users alike.

For example, we recently shipped an improvement to the official MCP rust-sdk repository.

We are making the foundations on which we are building stronger!

What is ContextVM?

A transport layer for MCP over the Nostr network. An open communication network built on cryptographic, censorship-resistant, and permissionless foundations!

https://contextvm.org/

Take a look at our recent demo on how to use our new LLM chat interface as a music research tool, by leveraging the WaveFunc MCP server.

Try it out for yourself!

https://nostr.at/nevent1qvzqqqqqqypzq6ehsrhjjuh885mshp9ru50842dwxjl5z2fcmnaan30k8v3pg9kgqythwumn8ghj7un9d3shjtnswf5k6ctv9ehx2ap0qy2hwumn8ghj7un9d3shjtnyv9kh2uewd9hj7qpqsvz0vz4g3ghvx6vttut462swy5v50dk5c5qv6gketrathnejfhcqsqa0cr

Our team is currently looking into CVM Bundles .cvmb, a packaging format to easily distribute servers and provide one-click deployment. The spec and implementation we are working on is inspired by .mcpb conventions by the official MCP working group
https://github.com/ContextVM/cvmi/pull/4

ContextVM Rust SDK 0.2.0 is out! 🚀

MCP over Nostr just got a whole lot more capable. Three big things landed today:

📦 Oversized transfers (CEP-22)
Until now, messages were capped by what fits in a single Nostr event (around 65 KB). Larger payloads, like big tool results or resources, now split into chunks automatically, get verified end-to-end, and reassemble on the other side. Servers only do this when the client supports it, so older clients keep working.

🌊 Open-ended streaming (CEP-41)
A server tool can now send a live stream of data back to the client while a request is still running, think progress, partial results, or log lines. The client reads it as an async stream, and the final normal response still arrives at the end. Streaming is negotiated automatically with clients that support it.

🌍 UniFFI bindings, write ContextVM in your language
The SDK now ships bindings for Python, Swift, Kotlin, and C/C++, all built on the same Rust core. That means native support on mobile, backend services, and scripting runtimes. There's a runnable examples/python/ folder to get you started.

Plus: smarter relay resolution (CEP-17), richer server announcements (CEP-6), optional human-friendly server profiles (CEP-23), progress-aware request timeouts, and an MSRV bump to 1.88.

Huge thanks to @aljaz for the job on uniffi bindings, and @Harsh for all the hard hork!

Ready to build on this? Read the docs and ship your first ContextVM server or client today: https://docs.contextvm.org/

And we have more news for today… hold on tight!

The latest issue of our CVM World is out, with so much news, this one is packed! 📦

If you like the work we’re doing, please consider subscribing to our biweekly newsletter. You could also contribute or donate some sats. Every little bit counts! 🫶

https://contextvm.substack.com/p/the-benefits-of-foss-the-contextvm

naddr1qvzq...

You can also use the SkateSpots server through the https://contextvm.org/chat interface 🛹

CEP-8 can now talk directly to AI agents. 🧵

With the just released ContextVM SDK 0.13.0 we're shipping full support for a new CEP-8 mode called Explicit Payment Gating.

Why we built it:
CEP-8 already let servers charge for their tools and capabilities, but only "behind the scenes", payment was handled silently by the transport layer. That's great for a wallet app. It's not great for an AI agent that actually needs to decide whether a call is worth paying for.

The fix:
Agents reason through normal MCP responses, so we turned payment into one. Now a paid call returns a clean "Payment Required" error (-32042) that carries human-readable instructions and one or more payment options. Feed that straight to your LLM, it reads the instruction, pays the invoice its own way, and retries the same call. No special SDK knowledge required, and it works with any MCP-compatible agent runtime.

This design is based on the MPP spec and inspired by HTTP's 402 Payment Required status code, a long-reserved error code for exactly this idea, now adapted for JSON-RPC and AI agents.

What it unlocks:
- AI agents that natively understand and handle paid MCP capabilities
- Per-call budgets and consent at the agent layer
- Payment as a first-class decision: pay, decline, or escalate to a human

Under the hood, each successful payment grants a single-use authorization that's matched to the exact same request on retry, and a "Payment Pending" error (-32043) keeps verification from spamming invoices.

Fully backward compatible, the existing silent "transparent" flow stays the default, so nothing breaks for current users.

Try it:
Upgrade to @contextvm/sdk 0.13.0 and turn on a paid capability with explicit gating.

Docs:
CEP-8 spec — https://docs.contextvm.org/reference/ceps/cep-8/
Explicit gating guide — https://docs.contextvm.org/how-to/payments/explicit-gating/
API reference — https://docs.contextvm.org/reference/ts-sdk/payments/explicit-gating/

Go build something agents can pay for. 🔌

#ContextVM #MCP #AI #Bitcoin #Lightning #Nostr

This is also a great example of what CVM allows, a full CVM server running in a browser tab! 🤯

ContextVM makes it easier to deploy your self-hosted infrastructure — No domains, no OAuth, no port forwarding!