RECOVER — Decide Safely After a Failed Machine Action

io.github.GSterlingPressPublished by GitHub account GSterlingPressv0.1.3Updated Nov 2, 2026

Choose the safest next action after an API, MCP, or agent failure before retrying.

VerifiedStreamable HTTPWeb executableDeveloper ToolsAI & ML

Overview

AI-generated overview

A remote decision service that turns a failed API, tool, or agent action into one of six safe next steps before you retry.

What you can do
  • Submit failure context (method, status, error, attempt, side-effect and idempotency flags) and get back a single deterministic decision.
  • Receive one of six outcomes: retry now, wait, repair the request, verify first, escalate, or abort.
  • Get a risk level, a reason, and a suggested next step alongside the decision.
  • Use it as a pre-retry checkpoint in agents, automations, and backend flows.
When to use it
Useful right after an API, MCP tool, or automation call fails and before an automatic retry, especially for side-effecting operations, ambiguous timeouts, 5xx or 429 responses, malformed requests, or exhausted retry budgets where a blind retry could duplicate an effect.
Requirements
A remote Streamable HTTP endpoint; no package, account, API key, or environment variable is declared. The assistant must be able to reach the hosted service over the network.
Before you install
Failure details you send (status, error text, attempt counts, side-effect flags) go to a third-party hosted service; the README says raw IPs and request payloads are not stored, but treat error messages as potentially sensitive. The tool only advises and does not act, so verify its recommendation before retrying side-effecting calls.

Installation

In SourceWeft

  1. Open RECOVER — Decide Safely After a Failed Machine Action in the dashboard and add it to a workspace.
  2. Enable the server for the chats that should use its tools.

Web executable via Streamable HTTP. Remote servers run from the web runtime once configured in a workspace.

Use with an agent

Paste this prompt into Claude Code, Codex, Cursor or another agent — or into SourceWeft chat. The agent reads this server's install guide, shows you its publisher, endpoint and sign-in, and adds it to your client once you agree. Keys never go into a command.

Read https://sourceweft.com/mcp/io.github.GSterlingPress%2Frecover/install.md and connect the MCP server it describes to my client. Before changing any config, show me its publisher, endpoint and sign-in method and wait for my OK. Never put an API key in a command or in a file I would commit; ask me where it should come from.

Read the install guide the agent follows

Other MCP clients

Add this to your client's mcpServers config.

{
  "mcpServers": {
    "recover-decide-safely-after-a-failed-mac": {
      "type": "http",
      "url": "https://recover-api-production-bbba.up.railway.app/mcp"
    }
  }
}

README

RECOVER

A failed machine action happened. What should happen next?

RECOVER is a tiny deterministic decision API for AI agents, MCP clients, automations, and ordinary backend systems. It converts failure context into one of six actions:

RETRY_NOW · WAIT · REPAIR_REQUEST · VERIFY_FIRST · ESCALATE · ABORT

Use it immediately after a failed API/tool/action and before an automatic retry—especially for side-effecting operations, ambiguous timeouts, rate limits, malformed requests, exhausted retry budgets, and uncertain execution outcomes.

Live production

Base URL: https://recover-api-production-bbba.up.railway.app

MCP Streamable HTTP endpoint: https://recover-api-production-bbba.up.railway.app/mcp

Activity dashboard: https://recover-api-production-bbba.up.railway.app/activity

Safe demo: https://recover-api-production-bbba.up.railway.app/demo

REST

POST https://recover-api-production-bbba.up.railway.app/v1/recover

json
{  "method":"POST",  "status":503,  "error":"upstream timeout",  "attempt":1,  "sideEffect":true,  "idempotencyKey":false}

Example response:

json
{  "decision":"VERIFY_FIRST",  "retry":false,  "risk":"high",  "reason":"A side-effecting non-idempotent action may have completed before the failure became visible.",  "next":"Verify external state or obtain an idempotency key before retrying."}

MCP

Tool: recover_decide

Point an MCP client at:

https://recover-api-production-bbba.up.railway.app/mcp

The server is designed for machine discovery and includes server.json metadata for the official MCP Registry.

Why call RECOVER?

Retry libraries answer how to retry. RECOVER answers the more important question first: what is the safest next action after this specific failure?

Good call sites include API clients, AI agents, browser automations, workflow engines, MCP hosts, job runners, payment/order flows, and any system where blindly retrying can duplicate a side effect or amplify a failure.

Analytics

GET /activity displays durable, privacy-safe usage telemetry and a permanent FIRST 10 REAL-WORLD STRANGER CONVERSIONS scoreboard. Real-use events are stored separately so discovery traffic cannot push milestones out of history. Raw IPs and request payloads are not stored.

Deploy

The repo is Railway-ready with Dockerfile and railway.toml. Mount a persistent volume at /data so the permanent real-use ledger survives restarts.

Source: README.md at commit 9409191

Tools

1

recover_decide

Use immediately after an API, MCP tool, automation, or agent action fails and before retrying. Especially use after timeouts, 5xx/429 responses, uncertain execution, or any side-effecting action that may already have completed. Returns the safest deterministic next action: RETRY_NOW, WAIT, REPAIR_REQUEST, VERIFY_FIRST, ESCALATE, or ABORT. Do not use before an action or for successful actions.

ParameterserrormethodstatusattemptheaderssideEffectmaxAttemptsidempotencyKeyproviderRetryableuncertainExecution

Version history

1
  1. v0.1.3LatestSep 16, 2026

Score

90% of the applicable checks met

How complete this listing is, computed from the registry entry, the endpoint and public sources. Not a rating of the server and not a statement that it is safe or working today: reachability is a dated check, and no installs or reviews go into the score.

  • Reached in the last 30 daysRequired25 pts
  • Auth declaredRequired10 pts
  • Tools listedRequired15 pts
  • DescriptionRequired15 pts
  • Logo10 pts
  • Documentation15 pts
  • License10 pts

    Declare a license in the repository or the registry entry.