Prototype · Preview: every server in the official registry, synced every 30 minutes. Sign in with Google to write reviews.
PopNx MCPMCP servers for people who use AI · explained in plain words

Deadlatch

by Diabloluoversion 0.1.2i★ 0iRuns on your computer
Registry nameio.github.Diabloluo/deadlatch

How to install

What it does

Local advisory pre-trade risk gate for agents. Never places orders.

  • Connects Runs on your computer
  • Account or key No key needed
  • Category Online shops & payments

Install

Pick the AI app you use, then copy the settings.

Passed 3/3 checks · Runs on your computer

First: which AI do you use?

Pick your app and the steps appear below.

Works with Claude, with some setup first

Install Python (uv) on your computer first (one time) · Only the Claude desktop app (Windows or Mac), not the web or phone

  1. Install Python (uv)

    One time only; it works for every server that needs Python (uv)

    Open the Python (uv) download page ↗

  2. Open the Claude desktop app’s settings file

    SettingsDeveloperEdit Config

  3. Paste these settings and save the file

    {
      "mcpServers": {
        "deadlatch": {
          "command": "uvx",
          "args": [
            "deadlatch",
            "--policy",
            "your-value",
            "--portfolio",
            "your-value"
          ]
        }
      }
    }
  4. Quit Claude and open it again

  5. Try it

    Use Deadlatch and tell me what you can do for me, then show me one thing

    ✓ If Claude says it is using Deadlatch or shows a Deadlatch tool in the answer, you are connected

Menu names checked against Anthropic’s help page on Nov 11, 2026 · your screen may differ slightly by app version

ChatGPT cannot connect Deadlatch yet

Deadlatch runs on your computer, but ChatGPT can only connect servers that run on the web

What you can do instead

A · Use an app that can connect Deadlatch

  • Install Python (uv) on your computer first (one time) · Only the Claude desktop app (Windows or Mac), not the web or phone
  • Install Python (uv) on your computer first (one time)
  • Install Python (uv) on your computer first (one time) · No one-click button: paste the settings and fill in the values yourself
  • Install Python (uv) on your computer first (one time) · No one-click button: paste the settings and fill in the values yourself

B · Others in Online shops & payments that work with ChatGPT

Picked from the most used in the same category, only ones that connect by address with no key

Works with Claude Code, with some setup first

Install Python (uv) on your computer first (one time)

  1. Install Python (uv)

    One time only; it works for every server that needs Python (uv)

    Open the Python (uv) download page ↗

  2. Open Terminal

    Mac: press Command+Space and type Terminal · Windows: open PowerShell

  3. Paste this command and press Enter

    claude mcp add --scope user deadlatch -- uvx deadlatch --policy your-value --portfolio your-value
  4. Type /mcp in Claude Code

    You should see Deadlatch listed as connected

  5. Try it

    Use Deadlatch and tell me what you can do for me, then show me one thing

    ✓ If Claude Code says it is using Deadlatch or shows a Deadlatch tool in the answer, you are connected

Menu names checked against Anthropic’s help page on Nov 11, 2026 · your screen may differ slightly by app version

Works with Cursor, with some setup first

Install Python (uv) on your computer first (one time) · No one-click button: paste the settings and fill in the values yourself

  1. Install Python (uv)

    One time only; it works for every server that needs Python (uv)

    Open the Python (uv) download page ↗

  2. Open Cursor’s MCP settings file

    ~/.cursor/mcp.json

    Paste these settings, fill in the empty values and save

    {
      "mcpServers": {
        "deadlatch": {
          "command": "uvx",
          "args": [
            "deadlatch",
            "--policy",
            "your-value",
            "--portfolio",
            "your-value"
          ]
        }
      }
    }
  3. Open the AI chat in Cursor in Agent mode

  4. Try it

    Use Deadlatch and tell me what you can do for me, then show me one thing

    ✓ If Cursor says it is using Deadlatch or shows a Deadlatch tool in the answer, you are connected

Menu names checked against Cursor’s help page on Nov 11, 2026 · your screen may differ slightly by app version

Works with VS Code, with some setup first

Install Python (uv) on your computer first (one time) · No one-click button: paste the settings and fill in the values yourself

  1. Install Python (uv)

    One time only; it works for every server that needs Python (uv)

    Open the Python (uv) download page ↗

  2. Open VS Code’s MCP settings file

    MCP: Open User Configuration

    Paste these settings, fill in the empty values and save

    {
      "servers": {
        "deadlatch": {
          "type": "stdio",
          "command": "uvx",
          "args": [
            "deadlatch",
            "--policy",
            "your-value",
            "--portfolio",
            "your-value"
          ]
        }
      }
    }
  3. Open the AI chat in VS Code in Agent mode

  4. Try it

    Use Deadlatch and tell me what you can do for me, then show me one thing

    ✓ If VS Code says it is using Deadlatch or shows a Deadlatch tool in the answer, you are connected

Menu names checked against VS Code’s help page on Nov 11, 2026 · your screen may differ slightly by app version

Other apps that use MCP, such as Windsurf, Zed or Gemini CLI, each add servers differently. See how in the app list.

See every app that uses MCP →

JSON settings are under “For developers” below

Stuck? Ask people who use Deadlatch💬 Discussion →
For developers: Terminal commands and JSON settings

Installed on your computer

Claude Code

claude mcp add --scope user deadlatch -- uvx deadlatch --policy your-value --portfolio your-value

Claude Desktop (claude_desktop_config.json)

{
  "mcpServers": {
    "deadlatch": {
      "command": "uvx",
      "args": [
        "deadlatch",
        "--policy",
        "your-value",
        "--portfolio",
        "your-value"
      ]
    }
  }
}

Cursor (~/.cursor/mcp.json)

{
  "mcpServers": {
    "deadlatch": {
      "command": "uvx",
      "args": [
        "deadlatch",
        "--policy",
        "your-value",
        "--portfolio",
        "your-value"
      ]
    }
  }
}

VS Code (mcp.json)

{
  "servers": {
    "deadlatch": {
      "type": "stdio",
      "command": "uvx",
      "args": [
        "deadlatch",
        "--policy",
        "your-value",
        "--portfolio",
        "your-value"
      ]
    }
  }
}

These settings are generated from the maker’s registry entry. If the README says otherwise, follow the README.

At a glance

Downloads per weeki
31/wk
Connection
Runs on your computer
Latest versioni
0.1.2
Published to registryi
Sep 22, 2026
Code last changedi
Oct 16, 2026
Maker
Diabloluo
Packages
pypi: deadlatch
Website
github.com
Registry status
Active

Which apps can use it

Based on how this server connects and on each app’s own docs, checked Nov 10, 2026.

Run on your computer in

ClaudeGooseLM StudioMsty StudioCherry StudioCursorVS Code + GitHub CopilotVisual StudioWindsurf (Devin Desktop)ZedKiroClineContinueJetBrains AI AssistantClaude CodeCodex CLIGemini CLIOpenCodenodecodeWarpLibreChatRaycast

All apps, with pros and cons →

Related guides

What it can do

Runs on your computer; the tool list appears in your AI app after you install it.

Basic checksi

We check information anyone can see; we have not read the code. Checks that do not apply are left out of the count. · Checked Nov 12, 2026

  • ✓ Pass

    Maker matches the repo

    The registry name was verified with the GitHub account Diabloluo, which owns the repo

  • ✓ Pass

    Still maintained

    Code last changed 27 days ago

  • ✓ Pass

    Has a licence

    Uses MIT

  • – Not applicable

    Endpoint responds

    This one runs on your computer; there is no internet endpoint

  • – Not applicable

    Sign-in with OAuth

    Runs on your computer and uses local settings instead

  • … Not checked yet

    No hidden instructions in tool descriptions

    Runs on your computer; we can only read tool lists from internet endpoints

Discussion

Ask, share a tip or report that it does not work. Sorted by what people marked helpful.

No posts yet. Be the first to ask or share.

Posts here are written by users, not the PopNx team.

Sign in with Google to post

User reviews

No reviews yet

Sign in with Google to write a review

Compared with others in Online shops & paymentsi

  • More stars than 0% of the other 2735 servers in this category
  • Published to the registry more recently than 35% of the other 2735 servers in this category
  • Basic checks equal to or better than 100% of the other 2735 servers in this category

Compared with every active server in this category in the registry. Stars count only when the publisher owns the repo.

Common questions

Is Deadlatch safe?

It passed 3 of 3 applicable basic checks. We have not read the code. Before adding a key or token, see what access the server asks for and use a key with only the permissions you need.

What keys or codes does it need?

The registry does not list a required key, though some servers ask you to sign in on first use.

Does it run on my computer or in the cloud?

It runs on your computer. Your AI app starts it when needed, and data does not pass through the maker’s servers (unless it calls an online service itself).

Is it still maintained?

The GitHub repo last changed 27 days ago, and the latest version (0.1.2) reached the registry 2 months ago.

Which apps can use it?

Apps that support MCP, such as Claude Desktop, Claude Code, Cursor and VS Code. You also need Python (uv) on your computer.

What can I use instead?

In the same category: Spree Commerce developer docs for AI agents, Stripe MCP, Sessy — Amazon SES observability

Others in the same category

Servers in the same category from other makers, sorted by stars.

Reading a card:↓ Downloads per weeki★ GitHub starsiThree pillars: reviews · checks · usageiUpdated = latest version published to the registryi

Spree Commerce developer docs for AI agents

Spree Commerce docs for AI coding agents: build headless B2B and marketplace ecommerce on its APIs.

by spreeWidely used★ 15.8K Stars for the whole repoUpdated 13 days ago

Runs in the cloud

Stripe MCP

Stripe’s official server lets AI view and manage customers, products, payments and invoices. Connect to Stripe’s cloud and sign in with OAuth.

by stripeWidely used★ 2.1KUpdated 13 months ago

Runs in the cloud✓ service owner · stripe.com

Sessy — Amazon SES observability

Read-only Amazon SES observability: search events, inspect bounces, pull delivery stats.

by marckohlbruggeGrowing★ 945Updated 3 months ago

Runs in the cloud

README from the maker

Open the README

Copied from the maker’s repo as written. PopNx MCP has not reviewed or endorsed it. HTML, images and scripts are removed. Links lead off this site.

Deadlatch

The pre-trade latch for your trading agent. Advisory-only.

Three things you need to know before anything else:

  1. You need an independent, cross-broker gate that you control. If an agent can place orders on your account, the last check between the agent and the broker should not be the agent itself — and it should not be locked to one broker's UI or rules.
  2. Deadlatch does not predict, does not recommend, and does not place orders. It answers one question only: *is this order allowed right now?* It is not a signal generator and it is not a broker.
  3. Every answer comes with reasons, evidence, and a local audit record on Linux/macOS. PASS / WARN / BLOCK is never a bare verdict — you can see which rule hit, why, and what was evaluated. Durable audit writes (init / append / repair / prune / shadow report) are supported on Linux and macOS only. On Windows the package still imports, and the rule engine / Guard / CLI / MCP still compute decisions; those write paths fail closed with audit_platform_unsupported. Offline verify / read of a stopped snapshot remain available. Windows MCP is a limited integration path, not a full trusted audit chain.

Honest boundary (please read): Deadlatch is advisory. It cannot force an agent that never calls it to call it, and it cannot stop an agent that ignores a BLOCK from submitting the order somewhere else. Whether the agent calls the guard and honors the result is the integrator's decision. Do not rely on this tool as a guarantee against loss — it is a gate, not an insurance policy.

Start here

The current install path is the verified PyPI deadlatch==0.1.2, published on 2026-09-22. The stable GitHub v0.1.2 Release targets b072e8a4d4b053e1d5a428dbb9c47e06652fbf4c; its wheel/sdist assets match the PyPI originals.

pip install deadlatch==0.1.2

The published MCP Registry name is io.github.Diabloluo/deadlatch at version 0.1.2, verified active/latest on 2026-09-22. The local server.json matches that published version. The command below starts the verified PyPI package directly.

MCP-first start:

uvx --from deadlatch==0.1.2 deadlatch-mcp --policy policy.yaml --portfolio portfolio.json

--policy and --portfolio are required local files. --audit-path and --kill-switch-path are optional. Use fictional or your own simulated inputs only. Deadlatch is advisory-only: it never places orders and cannot stop an agent that never calls it. Option orders must use the broker's unique full contract code as symbol. A historical GitHub pre-release remains at v0.1.0.dev1; it is not the current install path.

Then:

  1. Run a fictional-data Quick Start below (Python, CLI, or MCP) on Linux or macOS. Confirm PASS → BLOCK → local audit. Windows can import the package and run rule computation; it is not the audit-write Quick Start.
  2. Request a 20-minute integration assessment only if you already have an order-intent or simulated execution path: open the assessment form.

That GitHub issue is public. Do not paste accounts, positions, orders, API keys, tokens, customer names, or private paths. Security defects must go through GitHub Security Advisories, not a public issue.


Quick Start (60 seconds each)

All three quick starts use fictional data and a temporary audit path. They are executed from the same source scripts by the test suite, so they cannot drift from the documentation. Full audit-write / MCP Quick Start requires Linux or macOS. Windows may import the library and evaluate rules; it will not initialize or append a durable audit collection.

1. Python API
pip install dist/deadlatch-*.whl        # or: pip install -e .

python docs/quickstart/python.py

Shows Guard.from_policy(...) → Order / Portfolio → guard.check(...): a valid order returns PASS / 0; an oversized order returns BLOCK / 3 with the hit rules; on BLOCK the example caller stops — no broker call is ever made.

2. CLI
bash docs/quickstart/cli.sh                  # requires `deadlatch` on PATH

Creates fresh inputs in a temp directory (dynamic timestamps — never goes stale), then runs deadlatch check for PASS (exit 0), BLOCK (exit 3), an input error (exit 4), and a --json check, plus shadow report --json over the audit.

3. MCP (stdio)
python docs/quickstart/mcp_client.py         # requires `deadlatch` installed

Starts deadlatch-mcp as a real subprocess over stdio, lists the five tools, and calls check_order once for PASS and once for BLOCK. policy / portfolio / audit paths are server startup configuration — an agent cannot swap them as tool arguments. BLOCK is a constraint the caller must honor; technically the guard cannot force a fully bypassing agent to call it.


What it is / is not

Deadlatch isDeadlatch is not
A local, deterministic risk gate evaluated before you submitA signal generator, recommender, or portfolio optimizer
A library, a CLI, and a stdio MCP server — no broker connectivity, no policy mutationA broker adapter, an execution engine, or a market feed
An auditable check: every evaluation is written to a local JSONL logA cloud service, a database, or a telemetry sink
USD-only, single-leg orders, one snapshot per check (v0.1)Multi-leg, multi-currency, Greeks/IV-aware (see limitations)

Who should use it

  • Teams that already have an order-intent or simulated execution path and want an independent, deterministic pre-trade gate with a local audit trail.
  • Developers who want a small, dependency-light, fail-closed building block they can integrate into their own execution pipeline.
  • Anyone who wants to evaluate orders against a *policy they control*, expressed as plain YAML.

Who should not use it: anyone expecting a profit guarantee, a backtest engine, a portfolio manager, or a tool that enforces itself. If the agent never calls the guard, or ignores a BLOCK, nothing in this repository can stop it.

Core security boundary

  • Local: everything runs on your machine; no account credentials are ever stored, read, or transmitted.
  • No network core path: the library, CLI, and MCP server never open a socket, never register an HTTP/SSE route, and never call out for quotes or anything else (the MCP SDK's HTTP stack is a transitive dependency that business code never imports).
  • Never places orders: the core package (library, CLI, MCP server) has no broker connectivity and never submits orders. Experimental read-only mapping examples exist only in the development workspace; they are not included in the public candidate or the wheel, and they are not live-verified integrations.
  • Fail-closed: missing or malformed data → BLOCK (exit 3); input/config errors → exit 4; internal errors → exit 5. An uncertain state is never reported as PASS.
  • Direction is snapshot-derived: order-side text is never accepted as proof of a close. Stock and option closing intent is recognized only when a fresh portfolio snapshot contains a matching, opposite-side position with sufficient quantity. For options, symbol must be the broker's unique full contract code; never reuse an underlying ticker across different expiries, strikes, or rights.
  • USD-only (v0.1): any currency mismatch (order, portfolio, positions) is an input error (exit 4); the MCP account-status tool fail-closes on mismatch.

Write surface: the tool never modifies policy, portfolio, or kill-switch state, never connects to a broker, and never places an order. Two kinds of intentional local file writes exist:

  1. Audit subsystem: Guard.check() / check_order append one sanitized v2 record to a UTC day shard next to the logical audit path (30-day retention). Append is a bounded tail write under a shared lock; it does not rewrite history. A new collection needs one explicit deadlatch audit init (or initialize_audit_state) before append; Guard/MCP never auto-migrate. deadlatch shadow report and deadlatch audit prune delete whole shard files that fall outside the 30-calendar-day window. deadlatch audit repair --quarantine still uses lock/tmp files and os.replace when it isolates a truncated tail. Ordinary append is not crash-atomic across files. POSIX locks are cross-process. v0.1.2 durable audit writes are Linux/macOS only; Windows writers receive audit_platform_unsupported and do not skip directory fsync to report success.
  2. Explicit migration output: deadlatch migrate --output writes the migrated document only when you explicitly pass --output.

The public contract for these rules, Decimal thresholds, exit codes, and option-symbol requirements is docs/rules-spec.md.

The 12 rules (v0.1)

#RuleWhat it guards
R1kill_switchGlobal switch: off / full (block everything) / reduce_only (allow only inferred closing orders)
R2input_validityOrder passes schema, version gate, currency consistency, finite amounts (violations → exit 4)
R3max_order_quantitySingle-order quantity limit
R4max_order_valueSingle-order notional limit (options: price × multiplier × quantity)
R5max_symbol_exposureExposure per underlying (options by strike × multiplier × quantity)
R6max_total_exposurePortfolio gross exposure ratio
R7cash_margin_checkPost-trade cash floor and short-option margin
R8max_daily_lossDaily loss ratio (PnL / day-start equity)
R9max_drawdownDrawdown ratio from peak
R10order_time_validityOrder age / future timestamps (unparseable → fail-closed BLOCK)
R11data_freshnessPortfolio snapshot freshness (future snapshot → fail-closed)
R12missing_data_fail_closedMissing/null/ill-formed portfolio data → exit 3 (data unusable = risk)

Optional rules (R3–R7) are toggled by their config keys in policy.yaml; a missing optional key must be declared in acknowledged_disabled or the policy is rejected (exit 4). Mandatory rules (R1, R2, R8–R12) can never be disabled.

Exit codes

CodeMeaning
0PASS — the order is allowed as given
2WARN — proceed only if your execution policy explicitly allows warnings
3BLOCK — the order must not be submitted (risk rule or fail-closed data)
4Input / configuration error — the caller misused the API, not a risk event
5Internal / rule exception — treated as BLOCK (fail-closed)

In shadow mode the internal verdict is recorded (shadow_verdict) while the external projection is PASS / 0; kill-switch hits and exit 4/5 are never projected away.

Data contracts & migration

Schemas are versioned JSON Schema 2020-12 files shipped inside the package: order, portfolio, policy, result, audit-record, shadow-report, audit-maintenance-result, audit-prune-result, audit-write-state, audit-state-result. Explicit offline migration is available for legacy documents:

deadlatch migrate --kind order    --input order_v1.json    [--out

Only the beginning is shown; read the rest in the repo