RELATED 01
ClawMem
Local, SQLite-backed memory and hybrid search over your markdown for coding agents
211GitHub stars
190/wknpm downloads
48popularity score
Clawdi
Memories, past sessions, projects and encrypted vaults from a Clawdi account, as MCP tools
Official · Clawdi AI TypeScript Local · stdio Knowledge and memorySecurityAI and ML
55 popularity score ★ 106 ↓ 610/wk
Connect your client
Local (npm)
Add in one click
Opens the app, which asks before it adds the server.
Or add it by command or config
FirstTerminal
npm i -g clawdi
clawdi auth loginClaude CodeTerminal
claude mcp add clawdi -- clawdi mcpClaude Desktopclaude_desktop_config.json
{
"mcpServers": {
"clawdi": {
"command": "clawdi",
"args": [
"mcp"
]
}
}
}Cursor~/.cursor/mcp.json
{
"mcpServers": {
"clawdi": {
"command": "clawdi",
"args": [
"mcp"
]
}
}
}VS Code.vscode/mcp.json
{
"servers": {
"clawdi": {
"type": "stdio",
"command": "clawdi",
"args": [
"mcp"
]
}
}
}Codex~/.codex/config.toml
[mcp_servers.clawdi]
command = "clawdi"
args = ["mcp"]Gemini CLITerminal
gemini mcp add --scope user clawdi clawdi -- mcpSettings 3
CLAWDI_AUTH_TOKEN |
Optional: Clawdi token; takes precedence over the sign-in saved by `clawdi auth login` in ~/.clawdi/auth.json, so only needed without that login. |
|---|---|
CLAWDI_API_URL |
Optional: Clawdi API base URL, for a self-hosted backend; also settable with --api-url. |
CLAWDI_HOME |
Optional: Folder for config and sign-in instead of ~/.clawdi. |
What it does
Clawdi is a service for running AI agents in its cloud or connecting agents that run on your own computer, such as Claude Code, Codex, Hermes and OpenClaw. It keeps their shared context in one account. The `clawdi` npm package is mainly a CLI for signing in, syncing sessions and skills, and running a background sync daemon. It also contains an MCP server, started with `clawdi mcp`, which `clawdi setup` adds to supported agents for you.
The MCP server is a thin stdio proxy. It defines no tools of its own. It forwards `tools/list` and `tools/call` to the account's `/v1/mcp/clawdi` endpoint on the Clawdi API and passes back whatever the backend returns (packages/cli/src/mcp/server.ts). In the CLI 0.16.13 tag the backend registers 23 built-in tools (backend/app/routes/mcp_bridge.py). They cover durable memories (search, create, list, update, delete, extract from the conversation), synced sessions (search, list, read as Markdown, and public share links), projects, and vaults: list them, resolve `clawdi://` references to plaintext secrets, request values from the user through a one-time link, create vaults, and write or delete fields. Tools from apps connected in the Clawdi dashboard (Connectors, which the backend routes through Composio) are added to the same list. Tools whose permission scopes the signed-in token lacks are left out of the list.
The CLI waits up to 390 seconds for a tool call and 30 seconds for tool discovery. If a call times out it does not retry. It reports that the action may still have completed (server.ts).
What it needs
| Runtime | Node.js 24+ |
|---|
Node.js 24 or newer for the npm package. The native installer (install.sh / install.ps1) needs no Node. You need a Clawdi account, signed in once with `clawdi auth login`, or a token given in CLAWDI_AUTH_TOKEN. Connector tools appear only after apps are connected in the Clawdi dashboard. Runs on macOS, Linux and Windows.
Limits
Nothing runs offline: every tool call goes to the Clawdi API, and the server exits at start with "Not signed in" when it finds no credentials. The tool list depends on the token's scopes and on which connectors the account has linked, so it can be shorter or longer than the 23 built-in tools. The docs state that vault encryption is managed by the server and that the service can decrypt stored values. It is not a zero-knowledge store. Connectors can only be managed in the dashboard, not through the CLI or MCP. Features differ by agent software and by run path (cloud or connected).
Who it suits
Suits people who already use Clawdi, or want one account that keeps memories, past sessions and secrets shared across several coding agents. With Claude Code, Codex and Hermes, `clawdi setup` configures the server automatically. It is not a standalone memory server: everything is stored in Clawdi's cloud, or in a self-hosted copy of its FastAPI backend and PostgreSQL. If you want memory kept only on your own machine with no account, pick a local memory server instead. Be aware that the vault tools can return plaintext secrets to the agent.
Tools 23
memory_search- Semantic search over stored memories; 10 results by default, up to 50
memory_create- Save a new durable memory with a category and tags
memory_list- List stored memories with their IDs
memory_update- Replace the text of one memory by ID
memory_delete- Delete one memory by ID
memory_extract- Propose memories from the current conversation and save the ones the user approves
session_search- Keyword search over past synced sessions
session_list- List recent sessions, filtered by time, agent type or project
session_get- Return a session or share link's conversation as Markdown
session_share_create- Publish a public snapshot link of a session
session_share_list- List active session share links
session_share_revoke- Turn off a session share link
project_current_get- Return the project the caller is bound to
project_list- List projects the caller can read
project_get- Read metadata for one project by UUID
vault_list- List vaults attached to visible projects, without secret values
vault_get- List key names and clawdi:// references in one vault
vault_resolve- Turn up to 100 clawdi:// references into plaintext secrets
vault_request_create- Create a one-time link where the user enters new vault values
vault_request_status- Check whether a vault request has been filled, expired or conflicted
vault_create- Create a vault and attach it to a project
vault_item_upsert- Create or replace fields in a vault
vault_item_delete- Delete named fields from a vault