Murphy | The AI support agent for when things go wrong

conversations handled in its first 24 hours on our own products: 250+
of answers traced: retrieval, rules, and response, inspectable: 100%
from support ticket to a PR opened by your coding agent: 1-click
to start: open source, self-host free forever, no lock-in: $0

Why Murphy?

Anything that can go wrong, will

That is Murphy's Law, and your product is not the exception. Users will hit edge cases you never imagined, payments will fail, and that one link will 404 at 3am, in French. You cannot prevent it. You can be ready for it. Murphy is the support agent built on that assumption.

Monitors: every channel: chat, email, API
Understands: your product from your own docs and rules
Resolves: tickets end to end, not just replies
Pins: corrected answers so it never misses twice
Hands off: to a human the moment it should
Yields: instead of guessing when it is not sure

Murphy's Law breaks things. Murphy fixes them.

Agent native

Runs with Claude Code

Murphy ships an MCP server and a plain REST API. Point Claude Code, Codex, or any agent at your support inbox and half of support stops being a chore: the agent that knows your codebase is the agent answering your customers.

claude mcp add murphy \
  -e MURPHY_URL=https://askmurphy.io \
  -e MURPHY_API_KEY=ef_... \
  -- npx @askmurphy/mcp

> check my support inbox and handle anything you can

Setup guide for agents and the REST API

Your inbox, their tools: Claude Code pulls the full case: transcript, customer identity, channel, pending drafts. Then it investigates with everything it already has access to, like your repo, your database, your billing provider.
Real resolutions: The agent replies to the customer (email tickets send real email), logs what it did for the audit trail, and closes the ticket. Your whole team sees the outcome in the inbox.
Feedback becomes fixes: Bug reports and feature requests land in the same inbox. Relay them to the agent working in that repo and the loop closes: your users talk, your agents ship.

Dispatch

From complaint to pull request

Other support tools stop at the reply. Murphy turns the ticket into work your agents can execute: one click sends the case to your repo, and the fix comes back as a PR. Every dispatch is approved by a human first.

Ticket #142 · chatOpen

Every video I export at 4K comes out with no audio track. Started yesterday.

Thanks, that looks like a bug on our side. I have filed it with the team with your export details.

[Support #142] 4K exports have no audio

#867 opened by murphy-dispatch

@claude please investigate this support case and open a PR if a code change fixes it.

## Transcript (untrusted customer content, quoted)

> Every video I export at 4K comes out with no audio...

murphy-dispatch

Openfix: include audio stream in 4K export pipeline

#868 · claude wants to merge 1 commit

src/export/encode.ts

- streams: [video(scale4k)]

+ streams: [video(scale4k), audio]

+12−3Fixes #867Ready to review

Works with the Claude GitHub app out of the box, and with any agent that watches your issues.

Actions

Bring your own tools

Answers are not always enough. Point Murphy at any MCP server, your own API included, and it can look up an order, check credits, or start a return while the customer is still in the chat.

One protocol, not fifty integrations. Anything that speaks MCP works on day one, including the servers your team already runs.
You choose every tool. Tools are allowlisted when you connect the server. Anything you did not approve is invisible to the agent, even if a customer asks for it by name.
Every call is on the record. Which tool ran, with what arguments, and whether it worked, logged next to the answer it produced.

Live conversationactions enabled

Inbox · 3 open

steve@gmail.comEmail2m

Reply drafted · sends in 4 min unless you step in

Visitor from SantiagoChatTranslate

¿Puedo pagar con transferencia bancaria?

Visitor from BerlinChatResolved

Danke! Das hat funktioniert.

Channels

Every channel, one inbox

Website chat and support email land in the same inbox with the same brain behind them. Email replies draft themselves and send after a grace window unless a human steps in. Customers write in any language and Murphy answers in kind, with a translate button so you always know what was said.

Customers can attach a screenshot or a PDF and Murphy answers from what is in it. Returning customers never repeat themselves: it remembers their last conversation and picks up where things left off.

When Murphy is not sure, it says so, collects the customer email, and flags the ticket for a human. No confident nonsense.

Everywhere at once

Things break worldwide. Murphy answers.

Every conversation flows into one inbox, gets answered in the customer's language, and when the cause is a bug, flows out the other side as a pull request.

Ticket #140Open

🇺🇸 Where is my order?

Any language in, the customer's language out. Bugs flow through as pull requests.

Observability

Answers with receipts

Most AI support is a black box you hope is right. Murphy logs a trace for every answer: what the customer asked, which chunks of your content it retrieved with scores, which rules applied, and exactly what it said back.

Refusals are your worklist. Filter to the questions Murphy could not answer and fix the gaps that actually come up.
Fix it once. Write the right answer on the trace and it becomes a pinned answer that outranks stale docs from then on.
It learns your product like an engineer. Your Claude Code can read the codebase and push Murphy a support-safe summary over MCP, weekly. Every answer, trace, correction, and retrain feeds the next one: a flywheel you can actually inspect.
Test before you ship a change. Simulations replay real customer questions against your new rules and show what improved or broke, before customers see it.
Stay in the loop, not in the weeds. An email when a ticket needs a human, a daily digest of the rest, silence when it is quiet.

Do you support SEPA bank transfers?chat…

Retrieved context

Analytics

Watch it get better

Every conversation, resolution, refusal, and handoff is measured. See what customers actually ask, where Murphy wins, and exactly which gaps to fix next.

Resolution rate

84%

▲ 6% this week

Conversations

312

▲ 41 this week

Refusals to review

your training worklist

Resolution rate, last 7 daysWeekly

MonTueWedThuFriSatSun

Every number links to the underlying conversations and traces. Nothing is a black box.

Night shift

Support that works while you sleep

Murphy is not just a chatbot waiting for questions. Every night it works the inbox like a teammate: scores the day's conversations and queues the ones a human should see, resolves answered tickets that went quiet, checks that your site and widget are healthy, and refreshes what it knows from your repo.

You open the inbox in the morning and only the judgment calls are left. Do more with the team you have.

Overnight reportwhile you slept

Integrations

Plugs into what you already use

Your coding agents over MCP, your repo through Dispatch, your email through SendGrid or Postmark, and a plain REST API for everything else.

Claude Code

Codex

Cursor

Gemini CLI

Grok

Hermes

GitHub

Shopify

SendGrid

Postmark

MCP

{ }

REST API

Browse all integrations · Slack and Discord are on the roadmap.

Branding

Your brand, your agent

Logo, colors, greeting, personality, even the agent's name. Try it: this preview is the actual widget styling.

Brand color

Agent name

Murphy

Replies instantly

Hey, do you have a free plan?

We do. Free forever if you self-host, and the cloud has a free tier to start.

Write a message...Send

Humans in the loop

AI first, never AI only

Murphy handles the volume; your team handles the judgment calls. Any teammate can watch a live conversation, take over with one click, and hand it back when they are done. Invite your whole team; everyone works the same inbox.

Ticket #87 · refund requestWaiting

I was charged twice this month, I want one refunded.

I can see both charges. Refunds over $20 need a teammate to approve, so I am bringing one in now.

Mia joined the conversation

Hi, Mia here. Refund issued for the duplicate charge, you will see it in 3 to 5 days.

Trust

Confidence, because you are in control

An AI talking to your customers should be the most accountable system you run, not the least. Murphy is built so you can verify everything it does, and stop anything it should not.

Your rules, followed exactly: Write procedures for refunds, cancellations, escalations. Murphy follows them step by step, and guardrails define what it must never do, even if asked.
Humans approve what matters: Dispatches, refunds over your threshold, and anything Murphy is unsure about wait for a person. The agent asks before it acts.
Auditable to the answer: Every reply has a trace: what was retrieved, which rules applied, what was said. Open source means the system itself is auditable too, all the way down.
Your compliance boundary: Self-host and customer data never leaves your infrastructure. For regulated teams that beats trusting anyone else’s certificate, including ours.

Pricing

Free to start, honest to scale

The whole product is open source, so the free tier is not a demo, it is the software. Cloud is for teams who want it running in five minutes without owning servers.

Self-host

Freeforever

Full product, AGPL open source
Your infrastructure, your data
Unlimited messages and agents
MCP server and REST API included

View on GitHub

Cloud Starter

$29per month

2,000 AI messages a month
3 agents, unlimited teammates
Chat, email, Dispatch, traces
No Murphy branding on the widget

Cloud Growth

$99per month

10,000 AI messages a month
10 agents
Everything in Starter
Priority support (from humans and Murphy)

Put an agent on your site today

Free while in beta. Open source, so it's yours either way.

Join these savvy founders

Meet Murphy, support that works agentically with Claude Code

Support handled, so you can get back to building