Bray is a Nostr server built for AI agents, letting assistants such as Claude, ChatGPT, Gemini and Cursor work with the Nostr network through the Model Context Protocol (MCP). It is aimed at developers and technically comfortable users who want an AI model to handle identity, posting, relays and trust checks without manual copying of keys or endpoints. The project is free, open source under the MIT license, and currently at version v3.11.0.
Its design centers on trust. Rather than exposing a raw list of Nostr operations, Bray groups 234 tools into areas such as identity, social, trust, relay, marketplace, privacy and encrypted access, so an agent can weigh who it is dealing with before it acts. The last code update landed on October 30, 2026, so this is an actively maintained project. Windows appears in the documentation as a platform, but Bray runs beside an AI client rather than as a typical Windows app, so test any Bray for Windows setup with a small read-only task first.
What the Model Context Protocol Means for Nostr
The Model Context Protocol is a standard that lets an AI model call external tools and read external data through a server. The AI app is the client, and the server exposes a set of named actions the model can choose from. Bray applies this idea to Nostr.
Nostr is a decentralized protocol where identity is a cryptographic keypair and messages travel through relays rather than a single company's servers. Those two pieces fit together well. A model that understands the MCP tool list can translate a plain request, such as checking recent posts from a contact, into the correct Nostr operations.
The trade-off is that the model now has a route into your identity. Understanding this structure helps you decide which actions to allow and which to keep manual.
Who Bray Is For and Realistic Use Cases
Bray suits developers building agents that need Nostr capabilities, and Nostr users who already work with an AI assistant every day. A developer might ask Cursor to inspect how a relay responds while writing an app. A power user might ask Claude to summarize activity from a few identities they follow.
Researchers exploring decentralized social networks can also use Bray to gather context without building their own scripts. Because the tool count is large, the assistant chooses which group of functions to use, which keeps simple questions simple.
Bray is less suited to someone who wants a visual feed, a notification bell and a compose button. Those people will usually be better served by a dedicated Nostr client from the alternatives list.
How Trust Signals Shape an Agent-Driven Workflow
On an open network, publishing costs almost nothing, so the signal-to-noise ratio can be poor. Trust signals help separate accounts that have a consistent history from accounts that appear only to add noise. An agent that checks these signals before replying or following can behave more carefully than one that acts on every result.
Trust is a judgment aid, not a guarantee. A profile can look credible and still be wrong, and a low score does not prove bad intent. Treat the trust tools as one input among several, especially when a decision affects money, reputation or private information.
A practical habit is to ask the assistant to explain why it considers a profile trustworthy before you approve any follow-up action. That turns the trust layer into something you can audit.
Key features
A broad toolset of 234 tools
Trust-aware interactions
Signer pairing without copying secrets
Works with major AI clients
Supported coins
The documentation mentions Polygon and the Lightning Network alongside Nostr itself, and lists Ledger among hardware wallets. Bray's core focus is Nostr identity and messaging, so these references describe where it connects rather than a list of tokens it manages.
How to install Bray on Windows
- 1
Find the official release
Open the GitHub Releases page for the project, linked from the source repository at github.com/forgesworn/bray, and select the latest tagged build, v3.11.0 at the time of writing. Download only from the official website or GitHub, and avoid unofficial mirrors.
- 2
Verify the file
Compare the file's SHA-256 checksum with the value published alongside the release, or check the signature if one is provided. If the values do not match, delete the file and do not run it.
- 3
Place the program in a stable folder
Follow the installation instructions in the repository README and put Bray in a folder you control, not a temporary download directory. Make sure the path is one your AI client can launch.
-
Register Bray with your MCP client
Add Bray to your MCP client's configuration file, restart the client, and confirm the tool list appears. Start with read-only requests before enabling anything that publishes content or sends messages.
How to use Bray
Once Bray is registered with a client, the workflow is conversational. The steps below assume you have an MCP-compatible assistant and a Nostr signer app such as Amber or nsec.app.
- Connect your signer. Let Bray start the pairing from your signer app rather than copying a bunker:// URI by hand.
- Ask for a read-only task first, such as summarizing recent activity from an identity you already follow.
- Review each tool call the assistant proposes, especially any action that publishes a note or sends an encrypted message.
- Use trust-related requests to check a profile before you engage with it or follow it.
- Remove Bray from your client's configuration when you no longer need it, so the assistant does not keep access you do not use.
Pros & cons
- Free and MIT-licensed, with source code available on GitHub
- 234 tools cover identity, social, trust, relay, marketplace, privacy and encrypted access
- Works with Claude, ChatGPT, Gemini, Cursor and other MCP clients
- Signer pairing can start from the tool, avoiding manual bunker URI copying
- Recent code update and a clearly numbered current version, v3.11.0
- Requires an MCP-compatible AI client, so it is not a standalone app
- Windows is mentioned in the documentation, but the experience is not a conventional desktop program
- A large tool count means more actions to review
- Less suited to people who want a visual Nostr feed
Is Bray safe?
Bray handles signing keys and encrypted messages, so treat it as a security-sensitive tool. Download only from the official website or the GitHub Releases page, and verify the SHA-256 checksum or signature before running anything. Never enter a seed phrase or your nsec into a third-party website, and be wary of any site that asks for a bunker:// string outside your own signer app. Because an AI agent can act through 234 tools, check proposed publishing or messaging actions before approving them. Check the latest version on the official website before updating.
- Official source only. Download from github.com/forgesworn/bray. Ignore ads and lookalike domains.
- Check the SHA-256. Compare the installer hash with the value the developer publishes before running it.
- Never share your seed phrase. No developer, support agent or website — including CryptApps — will ever ask for it.
Alternatives
Most programs in this catalog, including Amethyst, Vega, Nostrid, Notedeck and Prettygood, are Nostr clients that people use directly to read and post. Bray works differently. It is a server that an AI assistant calls, so you talk to the model rather than navigating a feed. Vega is a cross-platform desktop client and Nostrid offers Windows binaries, while Prettygood centers its interface on building a web of trust. If you want a familiar app, start with a client. If you want an assistant to operate Nostr on your behalf, Bray is the closer fit.
Specifications
- Price
- Free
- Type
- Open source
- License
- MIT
- Latest version
- v3.11.0
- Last update
- 30 Oct 2026
- Windows support
- Unverified (Windows mentioned in docs)
- Source code
- github.com/forgesworn/bray
Frequently asked questions
Yes. Bray is listed as free and released under the MIT license, which lets you use, study and modify the code under its terms. The source is on GitHub, so you can review what it does. Any costs from your AI provider are separate and depend on your own subscription or usage.
Windows is mentioned as a platform in the project documentation. Bray runs as a server beside an AI client, so install it, connect it to your assistant and test with a read-only request first. If the setup behaves unexpectedly, check the README and the GitHub issues before going further.
It can be, if you set it up carefully. Let your signer app handle pairing, download only from the official website or GitHub, verify the checksum, and never type a seed phrase or nsec into a third-party site. Review any action that publishes or sends messages before you approve it.
Bray works with any MCP client. The project names Claude, ChatGPT, Gemini and Cursor. You add the server to the client's configuration file, restart the client, and confirm the tools appear before asking for Nostr tasks.
The Model Context Protocol is a standard that lets AI models use external tools and data through a server. Bray is an MCP server, so it exposes Nostr actions as named tools the assistant can call. This means one server can serve several different AI clients.
Some comfort with configuration files helps. You need to edit your MCP client's settings and verify a download with a checksum. Once setup is done, you do not need to write code to ask your assistant for Nostr tasks in plain language.
A regular Nostr app has its own screens for reading, posting and managing contacts. Bray has none of that interface. It gives an AI assistant the tools to perform those actions for you, so the interaction happens through the assistant rather than through buttons.
Verdict
Bray suits developers and technically confident Nostr users who want an AI assistant to work with the network, not a conventional app with feeds and buttons. Its free, MIT-licensed code, 234 tools and trust-focused design make it a serious option for that job. Because it touches signing keys and can act through an assistant, it rewards careful setup: start read-only, review actions, and keep your secrets inside your signer app. If you mainly want a visual client, a desktop program from the alternatives will fit better.