nostr-ops-mcp
A MCP server that provides NOSTR protocol tools for LLM agents, enabling signing, publishing, querying, and encrypted DMs with a defense-in-depth safety stack.
README
nostr-ops-mcp
A NOSTR identity for your LLM agent. MCP server that exposes NOSTR protocol primitives — sign, publish, query, NIP-19 encode/decode, NIP-05 lookup, encrypted DMs — as tools your agent can call. Drop it into Claude Desktop, Claude Code, Cursor, or any MCP-speaking client. Hand the agent a NIP-46 bunker key (not a raw nsec). Set a kind allowlist. Let it post on your behalf within rails you control.
v0.1 — full read + write + DM surface. 16 tools wrapped in a defense-in-depth safety stack: kind allowlist (deny-by-default), recipient allowlist, rate limits, optional two-step confirmation, structured audit log. Supports both nsec (dev) and NIP-46 bunker (production).
What you can do with this
- A bot that publishes kind:1 notes from your npub — daily summaries, scheduled posts, programmatic reactions to incoming events.
- A NOSTR sales agent — pair with
marketplace-mcpto publish NIP-15 stalls + products as the same identity. - Profile management —
nostr_publish_metadatafor kind:0 (always demands confirmation — overwriting your profile is irreversible without older relay data). - DM-driven workflows — a storefront agent that watches incoming DMs (with
nostr_list_dms), decrypts orders, and replies vianostr_send_dm. Default-off behindNOSTR_DM_TOOLS_ENABLED. - Cross-server identity — share the same NIP-46 bunker URI across
nostr-ops-mcpandmarketplace-mcp. One key, one identity, two specialized tool surfaces.
The safety stack is the load-bearing reason this is usable in production: an agent with the keys to publish as you can ruin your reputation in seconds if unconstrained. The server enforces what kinds it'll sign, what rate, optional second-step confirmation, and writes every call to a structured audit log.
The sixteen tools
Read-only — local (no network, no signer needed)
| Tool | Purpose |
|---|---|
nostr_decode |
Parse NIP-19 strings (npub / nsec / note / nevent / naddr / nprofile). nsec decoding gated behind NOSTR_ALLOW_NSEC_DECODE=true. |
nostr_encode |
Build NIP-19 strings from raw fields. nsec encoding deliberately not supported. |
Read-only — network (no signer needed for query/profile/nip05)
| Tool | Purpose |
|---|---|
nostr_get_pubkey |
Returns the signer's pubkey + npub. Errors clearly if no signer configured. |
nostr_list_relays |
Configured relay pool + each relay's connection status. |
nostr_query_events |
The workhorse. NIP-01 filters: kinds, authors, e_tag / p_tag / d_tag / t_tag tag filters (mapped to NIP-01 #e/#p/#d/#t internally — renamed in v0.2.0 for hosted-API schema compatibility), since / until / limit. |
nostr_get_profile |
Fetch + parse kind:0 metadata for a pubkey or npub. Returns the parsed JSON content (name, about, picture, nip05, lud16, …). |
nostr_verify_nip05 |
Resolve name@domain → pubkey via /.well-known/nostr.json. Optional expected_pubkey for verification mode. |
Write (require signer; gated by KindAllowlist + RateLimiter + optional confirm)
| Tool | Purpose |
|---|---|
nostr_publish_event |
The primitive write tool. Pass kind / content / tags. |
nostr_publish_text_note |
Convenience for kind:1. Reply/mention/hashtag shortcuts auto-assemble into NIP-10 tags. |
nostr_publish_metadata |
Kind:0 profile. Always demands two-step confirmation regardless of NOSTR_REQUIRE_CONFIRM — overwriting your profile is hard to reason about. |
nostr_publish_addressable_event |
Kinds 30000–39999 (replaceable). Sets the d tag automatically. The bridge marketplace-mcp uses for NIP-15. |
nostr_delete_event |
NIP-09 kind:5 soft delete. Best-effort — relays may ignore. |
nostr_confirm_publish |
Execute a token-gated publish. Single-use; safety pipeline re-runs. |
DMs (highest-risk; default-off via NOSTR_DM_TOOLS_ENABLED=true)
| Tool | Purpose |
|---|---|
nostr_send_dm |
NIP-04 (kind:4) DM with NIP-44 encryption by default; NIP-04 supported for legacy compat. Gated by NOSTR_DM_ALLOWLIST. |
nostr_list_dms |
Fetch + decrypt the thread with a counterparty. Auto-detects NIP-44 vs NIP-04 per event. |
nostr_decrypt_dm |
Decrypt a single ciphertext (when you already have the event from elsewhere). |
NIP-17 sealed/gift-wrapped DMs are not yet supported — deferred to a future v0.2 (rumor events + gift-wrapping add nontrivial complexity).
Requirements
- Node 20+
- A NOSTR signer — strongly preferred: a NIP-46 bunker URI from Amber (Android), nsec.app (web), or any other NIP-46 implementation. Legacy path: a raw nsec in
.env. The server logs a stderr warning at startup when nsec-on-disk is detected.
Install
# From npm (once published)
npx -y nostr-ops-mcp
# From source
git clone <repo>
cd nostr-ops-mcp
corepack enable pnpm
pnpm install
pnpm build
Configure
cp .env.example .env
# edit .env: set NOSTR_NIP46_URI (recommended) OR NOSTR_PRIVATE_KEY
# set NOSTR_RELAYS (comma-separated wss://)
# set NOSTR_ALLOWED_KINDS (required when a signer is configured)
The server auto-loads .env from this binary's own directory (next to dist/) — deliberately NOT from cwd, to avoid env-var collision when multiple MCP servers run in the same Claude Code session.
Required
| Var | Purpose |
|---|---|
NOSTR_RELAYS |
Comma-separated wss:// relays. Server refuses to start if empty. |
NOSTR_ALLOWED_KINDS |
Comma-separated event-kind numbers the server may sign. Required when a signer is configured. Example: 1,30017,30018 for text-note + NIP-15 marketplace. Default omits kind:0 (profile) and kind:5 (delete) — both easy to misuse. |
Signer — provide AT MOST one
| Var | Purpose |
|---|---|
NOSTR_NIP46_URI |
bunker://<pubkey>?relay=...&secret=... from Amber / nsec.app / Alby Account / any NIP-46 bunker. Recommended. |
NOSTR_PRIVATE_KEY |
Raw nsec1.... Dev/legacy only. Server warns at startup. |
Optional safety knobs
| Var | Default | Purpose |
|---|---|---|
NOSTR_READ_ONLY |
false |
Force read-only — disables all write tools. |
NOSTR_DM_TOOLS_ENABLED |
false |
Opt-in for send_dm / list_dms / decrypt_dm. |
NOSTR_DM_ALLOWLIST |
unset | Hex pubkeys allowed as DM recipients. Empty = NOSTR_DM_TOOLS_ENABLED alone gates. |
NOSTR_REQUIRE_CONFIRM |
false |
Two-step confirm: write tools return a token, nostr_confirm_publish executes. |
NOSTR_MAX_EVENTS_PER_MINUTE |
10 |
Rolling 60s rate limit on writes. |
NOSTR_MAX_DMS_PER_MINUTE |
5 |
Same but for DMs. |
NOSTR_ALLOW_NSEC_DECODE |
false |
Allow nostr_decode to return raw private key material. Don't enable unless you really need it. |
NOSTR_LOG_PATH |
./nostr-mcp.log |
Server log path. |
NOSTR_AUDIT_PATH |
./nostr-mcp-audit.log |
Structured audit log (one JSON line per tool call). |
Wire into an MCP client
Claude Code (project-scoped)
claude mcp add nostr-ops -s project node "$(pwd)/dist/index.js"
Claude Desktop / Cursor / other clients
{
"mcpServers": {
"nostr-ops": {
"command": "npx",
"args": ["-y", "nostr-ops-mcp"],
"env": {}
}
}
}
Because the server loads its own .env, leave the env block empty in the client config — keep secrets out of any committed file.
Safety model
Every write tool runs the pipeline in this order:
NOSTR_READ_ONLYgate — refuse outright.- Signer presence — refuse if neither nsec nor NIP-46 URI is configured.
- KindAllowlist — refuse if the event kind isn't in
NOSTR_ALLOWED_KINDS. - RateLimiter — refuse if the rolling 60s
eventsbucket is full. - Confirm gate — if
NOSTR_REQUIRE_CONFIRM=true(or the tool always-confirms, likepublish_metadata), return a 16-byte hex token instead of signing. - Sign + publish — via NDK; the signer handshake completes lazily on first use (relevant for NIP-46 where the bunker handshake is async).
- Audit log — append-only JSON line for every attempt (ok / blocked / error).
DM tools add three more checks on top: NOSTR_DM_TOOLS_ENABLED, DmAllowlist (per-recipient), and a separate dms rate bucket.
The floor is your signer. If using NIP-46, the bunker can refuse any sign request — that's the strongest safety boundary. This server's checks are belt-and-suspenders on top.
Verifying calls actually went through
tail -n 5 nostr-mcp-audit.log
Successful publish: {"ts":"...","tool":"nostr_publish_text_note","outcome":"ok","result":{"event_id":"...","relays_accepted":[...]}}. Blocked / error lines are equally structured. The audit log is append-only by intent — rotate it as part of your operational hygiene.
Testing
pnpm typecheck # tsc --noEmit
pnpm test # 13 vitest cases (KindAllowlist, RateLimiter, nip19 roundtrip)
pnpm build # dist/index.js (~58 KB ESM bundle)
For end-to-end testing against live relays, configure a throwaway nsec + a couple of public relays (damus.io, nos.lol) and run a small loop: nostr_publish_text_note → nostr_query_events to confirm the note round-tripped. The nostr_send_dm → nostr_list_dms loop validates the DM path (you can DM yourself for a closed-loop check).
Companion servers
nwc-mcp— Lightning wallet over NWC. Pair these to build sats-spending NOSTR agents.marketplace-mcp— NIP-15 marketplace publish (Shopstr-compatible). Uses the same signer setup as this server.
License
MIT — see LICENSE.
Contact / Issues
Built by LLMOps.Pro.
- NOSTR:
npub1hdg932jvwc3jdvkqywgqv0ue4nn60exrf92asy8mtazt3hjg7d2s2yw0nw— follow, DM, zap. - Lightning Address:
sovereigncitizens@getalby.com— for support zaps and "this was useful" tips. - Bug reports / feature requests: open a GitHub issue (link forthcoming).
- Security issues: please disclose privately via NOSTR DM before opening a public issue.
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。