writ runs on your own machine — the client starts it, so there is no endpoint to ping. 179 installs a week from pypi. Last commit 5 Oct 2026.
Commit-time policy checks for AI agent writes (ALLOW/DENY/STEP_UP) with a tamper-evident audit log.
We read the source, 6 d ago · rules ccca72095570
Found in continuous integration, deployment or infrastructure files, or in a neighbouring package of the same monorepo. None of this is installed when you add the server: it describes how the project is built and released. We list it because a leaked key in a build pipeline is still a real problem, but it is not something this server does on your machine.
proc = subprocess.Popen(
return sorted(p for p in root.rglob("*.py")
Is this your server and something here is wrong? Tell us — corrections are free and do not require a plan.
Today is the operative word: we check writ every 15 minutes and re-read its code on every release. Watch it and you find out the day that stops being true.
This server runs on your own machine — install it with the package manager and the client starts it for you. Package name taken from the official registry entry.
claude mcp add writ -- uvx writ-mcp
{
"mcpServers": {
"writ": {
"args": [
"writ-mcp"
],
"command": "uvx"
}
}
}
[mcp_servers.writ]
command = "uvx"
args = ["writ-mcp"]
{
"mcpServers": {
"writ": {
"args": [
"writ-mcp"
],
"command": "uvx"
}
}
}
{
"mcpServers": {
"writ": {
"args": [
"writ-mcp"
],
"command": "uvx"
}
}
}
This one needs environment variables set before it will start:
WRIT_API_KEY (Writ API key (get one free at https://withwrit.com)), WRIT_SPONSOR_TOKEN (Writ sponsor token, only required for sponsor tools (grant, revoke, reinstate)), WRIT_BASE_URL (Optional custom Writ API base URL (default: https://api.withwrit.com)).
The author declared them in the registry entry; get the values from the project itself.
Commit-time policy checks for AI agent writes (ALLOW/DENY/STEP_UP) with a tamper-evident audit log.
Pre-execution gate for AI-agent tool calls. Touch ID approval. Tamper-evident audit log.
Safe-by-default SQL guardrails for AI agents: AST-checked queries, per-table policy, audit log.
Tamper-evident audit logs for AI agents: submit, fetch, and verify cryptographic proof bundles.
BYOK Reddit/YouTube research CLI with a tamper-evident audit log and an MCP server.
Hosted CRM and task desk for AI agents: per-agent keys, scopes, pause and audit log.
Process monitoring for AI agents: start runs, submit per-task evidence, complete runs.
Nostr publishing, queries, and DMs for AI agents, with allowlists, rate limits, and an audit log.
Answers built from our own checks of this server.