Signet

Name: Signet
Author: Signet-AI

Local, shared memory for coding agents, with recall, a knowledge graph and cross-agent messaging over MCP

Official · Signet-AI JavaScript Local · stdio, Streamable HTTP Knowledge and memory Developer tools

GitHub npm Website

62 popularity score ★ 313 ↓ 1.1k/wk

Connect your client

Local (once it is running)

Add in one click

Add to Cursor Add to VS Code Add to LM Studio Add to Goose

Opens the app, which asks before it adds the server.

Or add it by command or config

Claude CodeTerminal

claude mcp add --transport http signet http://localhost:3850/mcp

Cursor~/.cursor/mcp.json

{
  "mcpServers": {
    "signet": {
      "url": "http://localhost:3850/mcp"
    }
  }
}

VS Code.vscode/mcp.json

{
  "servers": {
    "signet": {
      "type": "http",
      "url": "http://localhost:3850/mcp"
    }
  }
}

Codex~/.codex/config.toml

[mcp_servers.signet]
url = "http://localhost:3850/mcp"

Gemini CLITerminal

gemini mcp add --scope user --transport http signet http://localhost:3850/mcp

Local (npm)

Add in one click

Add to Cursor Add to VS Code Add to LM Studio

Opens the app, which asks before it adds the server.

Or add it by command or config

FirstTerminal

npm install -g signetai
signet setup
signet daemon start

Claude CodeTerminal

claude mcp add signet -- signet-mcp

Claude Desktopclaude_desktop_config.json

{
  "mcpServers": {
    "signet": {
      "command": "signet-mcp",
      "args": []
    }
  }
}

Cursor~/.cursor/mcp.json

{
  "mcpServers": {
    "signet": {
      "command": "signet-mcp",
      "args": []
    }
  }
}

VS Code.vscode/mcp.json

{
  "servers": {
    "signet": {
      "type": "stdio",
      "command": "signet-mcp",
      "args": []
    }
  }
}

Codex~/.codex/config.toml

[mcp_servers.signet]
command = "signet-mcp"
args = []

Gemini CLITerminal

gemini mcp add --scope user signet signet-mcp

Settings 4

`SIGNET_DAEMON_URL`	Optional: Address of the daemon that signet-mcp connects to (default http://localhost:3850).
`SIGNET_HOST`	Optional: Daemon host (default localhost).
`SIGNET_PORT`	Optional: Daemon port (default 3850).
`SIGNET_TELEMETRY_OPTOUT`	Optional: Set to 1 to stop the anonymous usage data sent by default.

What it does

Signet is a memory system for coding agents that runs as a local daemon. It takes in agent transcripts and connected sources such as Obsidian vaults, GitHub, Notion, Discord and imported documents. A background process it calls Dreaming turns that material into a knowledge graph of people, projects and claims, and every claim links back to its source. The MCP server is the on-demand side of Signet: the agent can search and store memories, browse the graph, explain where a claim came from, read old session transcripts, and message other agent sessions. Signet's hooks cover the automatic side (injecting context at session start, extracting after each prompt), and the two can run together.

There are two ways to connect. The `signet-mcp` program speaks stdio, and the daemon serves Streamable HTTP at http://localhost:3850/mcp. Either way, every tool is a thin wrapper that calls the daemon's HTTP API, so the daemon has to be running. Version 0.157.3 registers 57 tools in platform/daemon/src/mcp/tools.ts: 51 named tools plus six deprecated `code_*` aliases. The groups are memory, the knowledge graph, transcripts, cross-agent messages, secrets, GraphIQ code search and a router for other MCP servers. The README never lists these tools. Through `mcp_server_call`, and through proxy tools it creates at runtime, Signet can also pass calls on to external MCP servers you have installed in it, so the tool count can grow beyond 57.

What it needs

Runtime	Node.js

Install the Signet binary with `npm install -g signetai` (Node), `bun add -g signetai` or the install script. Runs on Linux x64/arm64, macOS x64/arm64, Windows x64 or Docker. Run `signet setup` once to create a workspace and pick an AI provider for Dreaming. Then start the daemon with `signet daemon start` before any client connects. For the GraphIQ code tools, enable the signet.graphiq plugin and index a project with `signet index <path>`.

Limits

Every tool fails without a running daemon. Over HTTP the daemon handles at most 8 MCP requests at once (503 beyond that) and accepts bodies up to 512 KiB. Calls to the daemon time out after 10 s by default. `memory_list` returns 100 rows by default, `session_search` 10 hits (max 20), `agent_message_inbox` at most 25 messages. `secret_exec` jobs run for 300 s by default, max 1800. The MCP docs also describe `agent_message_ack`, `signet_explain_claim`, `signet_code_clear` and `signet_code_briefing`, but v0.157.3 does not register them. Nightly builds may break. Slack, email, Telegram and WhatsApp sources are not yet available.

Who it suits

Signet suits developers who switch between Claude Code, Codex, Gemini CLI, OpenCode or Kimi Code and want one memory store on their own machine that all of them share, with a source behind every claim. It is a whole daemon with an AI provider behind it, not a small single-purpose server, so it is heavy if all you want is a notes file an agent can read. Teams can run one daemon as a shared server in team mode, where HTTP clients need a bearer token. If you don't want any tool to run shell commands, note that `secret_exec` can, so check what your client lets the agent call before you connect it.

Tools 51 of 57

memory_search: Searches stored memories by keyword and vector similarity, with filters
signet_recall: Explicit Signet recall, meant for Codex in place of the generic search
signet_source_search: Searches source-backed material: imported docs, Obsidian, Codex memory, transcripts
memory_store: Saves a new memory as evidence, with recall hints
signet_save_note: Writes a short note into Codex's native memory folder
memory_get: Fetches one memory by ID
memory_list: Lists memories, with paging and filters
memory_modify: Edits a memory and records a reason for the change
memory_forget: Marks a memory as forgotten (soft delete) with a reason
memory_feedback: Rates how useful injected memories were in this session
agent_peers: Lists active sessions of other agents
agent_message_send: Sends a message to another agent session, broadcasts to all, or relays over ACP
agent_message_inbox: Reads messages other agents have sent
secret_list: Lists the names of stored secrets, never their values
secret_exec: Queues a shell command with secrets injected as environment variables
secret_exec_status: Checks a queued secret_exec job and returns its redacted output
mcp_server_list: Lists the external MCP servers installed in Signet and their tools
mcp_server_search: Searches the tools of installed MCP servers
mcp_server_enable: Turns on an installed MCP server for the current scope
mcp_server_disable: Turns off an installed MCP server for the current scope
mcp_server_scope_get: Shows scope rules for installed MCP servers
mcp_server_scope_set: Sets which harness, workspace or channel an MCP server applies to
mcp_server_policy_get: Shows how routed tools are exposed to the agent
mcp_server_policy_set: Changes the tool exposure policy and expansion limits
mcp_server_call: Calls a tool on an installed external MCP server
session_bypass: Turns Signet's automatic hooks off or on for one session
knowledge_expand: Returns the graph context around one entity: aspects, attributes, related memories
knowledge_tree: Shows a compact outline of one entity's part of the graph
knowledge_list_entities: Lists top-level entities in the knowledge graph
knowledge_get_entity: Looks up one entity by name and summarises it
knowledge_list_aspects: Lists the aspects under an entity
knowledge_list_groups: Lists the groups inside an aspect
knowledge_list_claims: Lists the claim keys inside a group
knowledge_list_attributes: Lists the saved observations under one claim, current or all
knowledge_hygiene_report: Reports likely cleanup work in the graph, such as duplicates and missing fields
entity_list: Older name for knowledge_list_entities
entity_get: Older name for knowledge_get_entity
entity_aspects: Older name for knowledge_list_aspects
entity_groups: Older name for knowledge_list_groups
entity_claims: Older name for knowledge_list_claims
entity_attributes: Older name for knowledge_list_attributes
knowledge_expand_session: Returns session summaries that mention an entity
lcm_expand: Opens a node of the session history with its lineage and linked memories
session_search: Searches session transcripts, defaulting to a sub-agent's parent session
signet_session_search: Searches session transcripts only, apart from memory recall
signet_code_search: Searches the indexed project for symbols and code
signet_code_context: Reads a symbol's source and its surrounding structure
signet_code_blast: Works out what a change to a symbol would affect
signet_code_status: Shows the status of the GraphIQ index
signet_code_doctor: Checks the GraphIQ index for problems
signet_code_constants: Finds numeric and string constants shared across the project

Popularity

62 popularity score, on our list (the bar is 36)

★ 313 ↓ 1.1k/wk

GitHub stars, whole repository	313	24.4 points
Stars in 30 days	+9	5 points
npm downloads, last week	1,130	24.4 points
Official server	Yes	8 points

It lives in the folder dist/signetai of signet-ai/signetai, which holds more than this server. Its stars are the whole repository's, so in the score they never count for more than its own downloads.

Measured 2026-11-07. How the score works.

Where it is published

Code	signet-ai/signetai/dist/signetai ★ 313, the whole repository
npm	signetai 0.157.3, 2026-08-02

What it can reach

Runs	On your machine
Reaches	Files on your machine, Runs shell commands, The internet, Sends messages, Your code
Changes things	Yes: it can write, send, create or delete
Sign-in	None
Licence	Apache-2.0 (npm)
Last code change	2026-08-02 (a release)

Runs on your machine. signet-mcp relays calls to the local Signet daemon (localhost:3850), which keeps the memory store in your workspace. Tools write: they store, edit and soft-delete memories, write Codex memory notes, send messages between agents and change routed MCP server settings. `secret_exec` runs any shell command with stored secrets as environment variables and redacts their values from output. `mcp_server_call` forwards calls to other installed MCP servers. `agent_message_send` with via=acp posts to an ACP URL. Local mode (the default) needs no sign-in. Team mode requires a bearer token over HTTP, and hybrid mode requires one from hosts other than localhost. Anonymous telemetry is on by default and logged locally. SIGNET_TELEMETRY_OPTOUT=1 turns it off.

Over time

GitHub stars, whole repository 313 +9 (+3 %)

315304

9 Oct 20267 Nov 2026

npm downloads a week 1.1k −6.8k (−86 %)

28k294

2 Mar 20267 Nov 2026

Before the site measured it, read from npm's daily download counts.

Popularity score 62 +1 (+2 %)

6948

9 Oct 20267 Nov 2026

What is rising across the directory

Sources

Page checked 2026-10-14.

Badge for your README

Shows this server's rank in its category, and stars for a server that runs on your machine. It updates every hour.

MarkdownREADME.md

[![Signet on EveryMCPServer](https://everymcpserver.com/badge/signetai.svg)](https://everymcpserver.com/servers/signetai)

Connect your client

Add in one click

Or add it by command or config

Add in one click

Or add it by command or config

Settings 4

What it does

What it needs

Limits

Who it suits

Tools 51 of 57

Similar servers

State Memory MCP

marm-memory

Engram

SuperLocalMemory

agentmemory

Token Savior