A bridge holds memberships in two meshes and relays messages on a
single topic between them. Federation-lite without a broker-to-broker
protocol.
SDK additions:
- Bridge class (start, stop, EventEmitter for forwarded/dropped/error)
- MeshClient.joinTopic / leaveTopic / createTopic methods
- Loop prevention: plaintext hop counter prefix __cmh<n>: with maxHops
default 2; echo guard via senderPubkey == own session pubkey
CLI additions:
- claudemesh bridge run <config.yaml> long-lived process
- claudemesh bridge init prints config template
- Zero-dep YAML parser for the flat bridge config shape
The hop prefix is visible in message bodies — minor wart, fixed in
v0.3.0 by moving loop tracking into broker primitives.
SDK kept as devDependency since Bun bundles it into dist; no impact
on npm publish or runtime resolution.
Spec: .artifacts/specs/2026-05-02-v0.2.0-scope.md
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Issuance flow over WS for now (REST endpoints come next slice).
Plaintext secret returned ONCE on create — never recoverable.
- broker: 3 WS handlers (apikey_create/list/revoke), wire types in
union, audit log on issuance + revoke
- ws-client: apiKeyCreate/List/Revoke with resolver maps, response
dispatch
- CLI: claudemesh apikey create <label> [--cap a,b] [--topic c,d]
[--expires ISO]; list shows status, scope, last-used; revoke by id
- policy: apikey create + revoke prompt by default (issuing or
disabling a credential is meaningful)
Default capability set is "send,read" — least privilege for unscoped
keys (admin must explicitly opt-in).
Spec: .artifacts/specs/2026-05-02-v0.2.0-scope.md
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Broker (apps/broker/src/index.ts)
- Unified disconnect/kick handler uses close code 1000 for disconnect
(CLI auto-reconnects) vs 4001 for kick (CLI exits, no reconnect).
- Ban now closes with code 4002.
- Hello handler: revoked members get a specific 'revoked' error with a
'Contact the mesh owner to rejoin' message, then ws.close(4002).
Previously banned users saw the generic 'unauthorized' error.
- list_bans handler returns { name, pubkey, revokedAt } for each
revoked member.
CLI (apps/cli)
- ws-client: close codes 4001 and 4002 set .closed = true and stash
.terminalClose so callers can surface a friendly message instead of
the low-level 'ws terminal close' error. Revoked error in hello is
also captured as a terminal close.
- withMesh catches terminalClose and prints:
4001 → 'Kicked from this mesh. Run claudemesh to rejoin.'
4002 → the broker's 'Contact the mesh owner to rejoin.' message
- kick.ts now exports runDisconnect + runKick with clear hints:
'disconnect' → 'They will auto-reconnect within seconds.'
'kick' → 'They can rejoin anytime by running claudemesh.'
- cli.ts adds 'disconnect' dispatch; HELP updated.
Semantics:
disconnect: session reset, no DB state, auto-reconnects
kick : session ends, no DB state, user must manually rejoin
ban : session ends + revokedAt set, cannot rejoin until unban
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Promote CLI from 1.0.0-alpha.42 to stable 1.0.0 so
`npm i -g claudemesh-cli` installs the current release without
needing the @alpha dist-tag.
Both dist-tags now point at 1.0.0 — `@alpha` kept as an alias for
continuity so existing docs, install scripts, and scheduled upgrade
commands keep working.
upgrade + doctor commands updated to prefer the `latest` dist-tag
(falling back to `alpha`) and to suggest `npm i -g claudemesh-cli`
without the @alpha suffix.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Three bugs compounding when multiple peers share a display name:
1. list_peers (MCP + CLI) truncated pubkey to 12 hex chars with an
ellipsis. A truncated pubkey cannot be used as a routing key, so
the caller had no way to disambiguate visually.
2. send_message required the full 64-hex pubkey and refused prefix
input, forcing callers to rely on --json output to get a full key.
3. Name-based resolution returned the first exact match without
filtering the caller's own session — so "send to <my-own-name>"
would bounce against the broker's self-send guard when another
session of the same user was the intended target.
Fixes:
- list_peers now prints 16-char pubkey prefix labelled "pubkey: …"
(MCP) and appends it to CLI output
- send_message accepts any 8–64 hex-char prefix and resolves against
live peer lists across joined meshes; unique match routes, multi-
match returns a disambiguation error listing each candidate's
displayName + pubkey + cwd
- Name matches now skip the caller's own session pubkey; multiple
same-named matches fail loudly with a copy-pasteable pubkey
disambiguation hint instead of silently picking one
- Full 64-char pubkeys without a live match still queue at the
broker (preserves offline-delivery semantics)
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
If the arg isn't a URL and matches a mesh already in local config,
print a hint pointing at `launch --mesh <slug>` instead of treating
the slug as an invite code. Avoids the 501 invite_v2_disabled confusion
when users try to "enter" a mesh they already own.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The peers command opens its own WS to each mesh, which briefly appears
as a hostname-PID peer. Filter it out by session pubkey.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Broker WS handlers:
- kick: disconnect peer(s) by name, --stale duration, or --all.
Authz: owner or admin only. Closes WS + marks presence disconnected.
- ban: kick + set revokedAt on mesh.member. Hello already rejects
revoked members, so ban is instant and permanent until unban.
- unban: clear revokedAt. Peer can rejoin with their existing keypair.
- list_bans: return all revoked members for a mesh.
Session-id dedup (previous commit): handleHello disconnects ghost
presences with matching (meshId, sessionId) before inserting the new
one. Eliminates duplicate entries after broker restarts.
CLI (alpha.37):
- claudemesh kick <peer|--stale 30m|--all>
- claudemesh ban/unban <peer>
- claudemesh bans [--json]
- Uses new sendAndWait() on ws-client for request-response pattern
over WS (generic _reqId resolver).
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Broker (all need redeploy):
- sweepOrphanMessages: DELETE undelivered message_queue rows older
than 7 days; hourly sweep. Stops unbounded growth when a sender
typos a name (queued forever, never claimed).
- Per-member send rate limit: TokenBucket(60/min, burst 10) keyed on
memberId so reconnecting can't bypass. Surfaces as queued=false,
error='rate_limit: ...'.
- Pre-flight size cap: reject at handleSend if nonce+ciphertext+
targetSpec exceeds env.MAX_MESSAGE_BYTES with a clear error
instead of silent WSS frame-level kill.
- No-recipient reject: for direct sends, check any matching peer
is connected BEFORE queueing. Kills the self-send silent drop
(sending to your own pubkey when you only have one session
connected) and typo-to-offline-peer silent drops.
- WSAckMessage.error field added for structured failure reasons.
CLI:
- ws-client ack handler reads msg.queued and msg.error; surfaces
rate_limit / too_large / no_recipient to callers instead of
returning ok:true with a dummy messageId.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
When Alice's session-A encrypts a direct message to Bob (target = Bob's
stable member pubkey) and Bob's session-B receives it, Bob has BOTH an
ephemeral session secret key and the member secret key. The old code
only tried session_sk, then silently failed with '⚠ message from
<sender> failed to decrypt' even though the message was valid —
just encrypted to the member key.
Now: try session first, fall back to member on null. Matches the
sender side's choice freedom (encrypt using either key).
Repros when: user opens multiple Claude Code sessions (all use the
same member key but each generates its own session key), and one
session sends to another by display-name resolution which returns
the member pubkey.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Two bugs that combined to make Claude's peer-send look successful even
when the recipient didn't exist:
1. resolveClient fell through to 'let the broker try' when a single
mesh was joined and the name didn't match any peer. The broker
queued the message against the literal unknown string, matched no
peer in fan-out, but returned a messageId — so the CLI reported
'✓ lezg → msgId' for a peer that was never there.
Now: refuse to send, list the known peer names.
2. list_peers showed the same pubkey multiple times with different
display_names (one per live session) without hinting that they
were the same member — so Claude treated them as distinct people.
Now: annotate with '[shares key with N other session(s)]' so the
caller understands one pubkey = one identity.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- info/inbox commands → unified render.ts
- install route: drop in-memory counter, rely on PostHog + structured logs
- docs: roadmap, CLAUDE.md reflect alpha.31 state
- tests workflow now also builds + smoke-tests the CLI bundle
- homebrew tap bootstrap kit in packaging/homebrew-tap-bootstrap/
(README + copy of the formula template for dropping into the tap repo)
- upstream Claude Code issue draft for rich <channel> UI
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- apps/cli/ is now the canonical CLI (was apps/cli-v2/).
- apps/cli/ legacy v0 archived as branch 'legacy-cli-archive' and tag
'cli-v0-legacy-final' before deletion; git history preserves it too.
- .github/workflows/release-cli.yml paths updated.
- pnpm-lock.yaml regenerated.
Broker-side peer-grant enforcement (spec: 2026-04-15-per-peer-capabilities):
- 0020_peer-grants.sql adds peer_grants jsonb + GIN index on mesh.member.
- handleSend in broker fetches recipient grant maps once per send, drops
messages silently when sender lacks the required capability.
- POST /cli/mesh/:slug/grants to update from CLI; broker_messages_dropped_by_grant_total metric.
- CLI grant/revoke/block now mirror to broker via syncToBroker.
Auto-migrate on broker startup:
- apps/broker/src/migrate.ts runs drizzle migrate with pg_advisory_lock
before the HTTP server binds. Exits non-zero on failure so Coolify
healthcheck fails closed.
- Dockerfile copies packages/db/migrations into /app/migrations.
- postgres 3.4.5 added as direct broker dep.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Terminals spawned by `claudemesh launch` were dropping keystrokes at
claude's prompt and showing the launch wizard re-rendering on top of
claude's TUI. Two compounding causes:
1. spawn() + child.on('exit') kept the parent node event loop alive
during claude's lifetime. Any stray readline 'data' listener or
late render from the wizard could fire on the inherited stdin/
stdout, stealing keystrokes or painting over claude's Ink TUI.
2. Raw mode / alt-screen / hidden cursor set by the wizard helpers
was not reliably restored before the handoff.
Fix:
- Swap spawn for spawnSync so the parent event loop is fully blocked
while claude runs. No listener or setImmediate can fire during
claude's lifetime.
- Hard TTY reset right before the spawn: setRawMode(false),
removeAllListeners on stdin, show cursor (ESC[?25h), exit alt
screen (ESC[?1049l). Defensive — survives partial wizard cleanup.
- Move cleanup() registration to process.on('exit') so it runs
synchronously on every exit path (normal, signal, throw).
- Preserve signal forwarding: if claude dies from a signal, re-raise
the same signal on the parent so exit codes propagate correctly.
Bumps to v0.10.6.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Wires the v2 invite protocol end-to-end from a CLI user's perspective.
Broker foundation landed in c1fa3bc; this commit is the glue between
it and the human.
API (packages/api)
- createMyInvite now mints BOTH v1 token (legacy) AND v2 capability.
Two-phase insert: row first (to get invite.id), then UPDATE with
signed canonical bytes stored as JSON {canonical, signature} in the
capabilityV2 column. Broker's claim handler parses the same shape.
- canonicalInviteV2 locked to `v=2|mesh_id|invite_id|expires_at|role|
owner_pubkey_hex` — byte-identical to apps/broker/src/crypto.ts.
- brokerHttpBase() helper rewrites wss://host/ws → https://host for
server-to-server calls.
- POST /api/public/invites/:code/claim — thin proxy to broker;
passes status + body through, 502 broker_unreachable on fetch fail,
cache-control: no-store.
- POST /api/my/meshes/:id/invites/email — mints a normal v2 invite
via createMyInvite, records a pending_invite row, calls stubbed
sendEmailInvite (logs TODO for Postmark wiring in a later PR).
- New schemas: claimInviteInput/ResponseSchema,
createEmailInviteInput/ResponseSchema, v2 fields on
createMyInviteResponseSchema.
- v1 paths untouched — legacy /join/[token] and /api/public/invite/:token
continue to work throughout v0.1.x.
CLI (apps/cli)
- New `claudemesh join <code-or-url>` subcommand.
- Accepts bare code (abc12345), short URL (claudemesh.com/i/abc12345),
or legacy ic://join/<token>. Detects v2 vs v1 and dispatches.
- v2 path: generates fresh ephemeral x25519 keypair (separate from
the ed25519 identity) → POST /api/public/invites/:code/claim →
unseals sealed_root_key via crypto_box_seal_open → persists mesh
with inviteVersion: 2 and base64url rootKey to local config.
- Signature verification skipped with TODO — v0.1.x trusts broker;
seal-open is already authenticated.
- apps/cli/src/lib/invite-v2.ts: generateX25519Keypair, claimInviteV2,
parseV2InviteInput.
- state/config.ts: additive rootKey?/inviteVersion? fields.
CLI friction reducer
- apps/cli/src/index.ts: flag-first invocations
(`claudemesh --resume xxx`, `claudemesh -c`, `claudemesh -- --model
opus`) now route through `launch` automatically. Bare `claudemesh`
still shows welcome; known subcommands dispatch normally.
- Removes one word of cognitive load: users never type `launch`.
No schema changes. No new deps. v1 fully backward compatible.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Claude Code's MCP prompts path doesn't support the context field
natively. When a skill has context:"fork", prepend an instruction
telling the model to use the Agent tool with the specified agent
type and model.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Welcome was silently dropped when sent before Claude Code's
notifications/initialized. Add 2s delay after WS connects to
ensure the MCP handshake is complete.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Move startClients() to run after server.connect(), not before.
MCP server is available to Claude Code in <0.5s instead of ~30s.
Tool handlers gracefully return errors until WS is ready.
Push event wiring happens in background callback.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Claudemesh MCP server now declares prompts:{} and resources:{} capabilities.
Mesh skills auto-appear as /claudemesh:skill-name slash commands in Claude Code
via prompts/list+get, and as skill://claudemesh/{name} resources for the
upcoming MCP_SKILLS protocol. share_skill accepts optional metadata (when_to_use,
allowed_tools, model, context, agent) stored in the manifest jsonb column.
Change notifications sent on share/remove so Claude Code refreshes.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
System messages (watch_triggered, mcp_deployed, peer_joined, etc.)
have senderPubkey='system' with empty ciphertext. The push handler
now formats them as readable plaintext instead of failing to decrypt.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
System messages (watch_triggered, mcp_deployed, peer_joined, etc.)
have senderPubkey='system' with empty ciphertext. The push handler
now formats them as readable plaintext instead of failing to decrypt.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Runner /load now accepts gitUrl, npxPackage, or sourcePath. It handles
git clone and npm install internally. Broker no longer needs shared
volume for source extraction — just tells the runner what to fetch.
CLI mesh_mcp_deploy now supports npx_package as a third source type.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
claudemesh launch now supports:
--resume <id> / -r — resume a previous Claude Code session
--continue / -c — continue the most recent conversation
When resuming, skips generating a new session ID so the mesh peer
identity persists. The detectClaudeSessionId() fallback in ws/client.ts
picks up the existing session UUID from the .jsonl file.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
claudemesh launch now generates a UUID and passes it to claude via
--session-id flag + CLAUDEMESH_SESSION_ID env var. The MCP server
reads this and sends it in the hello handshake.
Fallback: when launched without claudemesh launch (e.g., claude --resume),
detectClaudeSessionId() scans ~/.claude/projects/ for the most recent
.jsonl file and extracts the session UUID from the filename.
Benefits:
- Broker detects reconnections (same session = restore state)
- Multiple peers in same project dir get unique identities
- Session identity persists across --resume
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
The Wave 3I handlers (vault_set, vault_list, vault_delete, mesh_mcp_deploy,
mesh_mcp_undeploy, mesh_mcp_update, mesh_mcp_logs, mesh_mcp_scope,
mesh_mcp_schema, mesh_mcp_catalog, mesh_skill_deploy) were lost during
the re-apply phase. Tools were registered in tools/list but returned
"Unknown tool" because the switch cases in server.ts were missing.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Add the foundation for deploying and managing MCP servers on the VPS
broker, with per-peer credential vaults and visibility scopes.
Architecture:
- One Docker container per mesh with a Node supervisor
- Each MCP server runs as a child process with its own stdio pipe
- claudemesh launch installs native MCP entries in ~/.claude.json
- Mid-session deploys fall back to svc__* dynamic tools + list_changed
New components:
- DB: mesh.service + mesh.vault_entry tables, mesh.skill extensions
- Broker: 19 wire protocol types, 11 message handlers, service catalog
in hello_ack with scope filtering, service-manager.ts (775 lines)
- CLI: 13 tool definitions, 12 WS client methods, tool call handlers,
startServiceProxy() for native MCP proxy mode
- Launch: catalog fetch, native MCP entry install, stale sweep, cleanup,
MCP_TIMEOUT=30s, MAX_MCP_OUTPUT_TOKENS=50k
Security: path sanitization on service names, column whitelist on
upsertService, returning()-based delete checks, vault E2E encryption.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Persistent MCP servers (opt-in via `persistent: true`) survive host
disconnects — they appear as offline in mcp_list and auto-restore when
the host reconnects. Ephemeral servers (default) still clean up on
disconnect. Offline servers return a clear error on mcp_call with
time-since-disconnect info.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Save groups, profile, visibility, summary, display name, and cumulative
stats to a new mesh.peer_state table on disconnect. On reconnect (same
meshId + memberId), restore them automatically — hello groups take
precedence over stored groups if provided. Broadcast peer_returned
system event with last-seen time and summary to other peers.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Peers now receive [system] notifications when MCP servers join or
leave the mesh, with tool names and hosting peer info.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Teaches AI when to use filesystem (local), read_peer_file (remote
<1MB), or share_file (persistent, no size limit).
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
