Build and deploy MCP
servers and apps

One SDK, one cloud, every client. Ship to ChatGPT and Claude, run for your own agents, and see every tool call in production.

Start deploying Book a call

AAcme Labs

/chart-serverproduction

SearchKARTLPK

Servers

chart-serverpostgres-mcpinternal-toolslegacy-bridge

chart-server

Livechart-server.mcpfy.app/mcp

OverviewToolsLogs

Tool calls

12,481+18%

p50 latency

38ms−6ms

Error rate

0.2%−0.4%

ToolCallsp95

chart_salesobject5,20488ms

query_dbobject3,918112ms

list_reportsarray2,44141ms

export_pdfobject918402ms

InspectorClaude Opus 5

→ tools/call

{
  "name": "chart_sales",
  "arguments": {
    "period": "Q3",
    "limit": 12
  }
}

← result20041ms

{
  "content": [{
    "type": "resource",
    "uri": "ui://chart"
  }]
}

Verified across

ChatGPTClaudeGeminiCursor

Trusted by teams building on Mcpfy.

emergenceSabadellSmileARGENISSWebSeodocfyncreagiaDifferentByteZerone

Distribution

One deploy. Works with every client.

Push once and the same server answers in ChatGPT, in Claude, and in your own agents. No forks, no per-client rewrites, no second pipeline.

AI Chats

ChatGPTClaudeGemini EnterpriseCopilot 365CopilotKit

Coding agents

CodexClaude CodeCursorVS Code

Internal agents

OpenAI AgentsClaude Agent SDKMastraLangChainVercel AI SDK

Start building

Live routing1 Build → all surfaces

Discovery

In the ChatGPT Apps Store.
In Claude Connectors. From one build.

Shipping is not the same as being found. Your server shows up as an entry people can browse and install, in the connector directory of a chat client and in the app store next to it, with the icon, copy and screenshots already filled in.

Connector directoriesApp storesSubmission pack

See it against a real server

Connectors

Connect the tools you already use. Your servers show up here.

Northwind CRMSearch accounts, update deals, and draft follow-ups.

Kestrel AnalyticsQuery events and chart them in the conversation.

Meridian DocsRead, summarise and edit team documents.

SundialFind a slot and put the meeting on the calendar.

Appsbeta

Chat with your favourite apps

Chart anything you store

chart-server · featured today

Aperture FilesStore, tag and retrieve assets›

Cogent SearchIndex anything, answer from it›

IroncladRedline and compare contracts›

BrightpathPlan routes and track delivery›

What you skip

Why teams pick Mcpfy

Seven jobs stand between a working tool and something people can install. You can write all of them, or none of them.

By hand

With Mcpfy

Hosting

Pick a host, write a Dockerfile, wire up TLS and a domain.

Connect the repository. Every push ships.

Preview environments

Stand up a second stack, or test in production and hope.

A live MCP URL for every branch, automatically.

Secrets

A .env file, a secret manager, and a way to keep them in sync.

Per-environment variables injected at runtime.

Debugging

console.log, then redeploy, then read the logs again.

Every JSON-RPC frame captured, with session replay.

Testing across clients

Open ChatGPT, then Claude, then Cursor, and click through each one.

One suite, run across clients and models, judged automatically.

Analytics

Bolt on a generic APM that has never heard of a tool call.

MCP-native volume, latency, errors and per-tool health.

Getting listed

Read the directory guidelines and assemble the pack by hand.

Publish checks, then a generated submission pack.

The pipeline

From first commit to production

Four stages, one platform. No second vendor, no CI plugin, no weekend of YAML.

1build

Start from an SDK, a skill, or an API spec.

Write the server yourself with the Mcpfy SDK, or skip the boilerplate and let a skill in your coding agent scaffold it for you. Have an OpenAPI spec lying around? Hand it over and get typed tools back, no code required.

SDK

One command, a running server.

Read the docs

Skill

Your coding agent scaffolds it for you.

Get the skill

API Spec

Hand over OpenAPI, get typed tools back.

Import a spec

acme-mcp · zsh

➜ npx create-mcpfy-app

✓ tools/chart-sales.ts

✓ widgets/Chart.tsx

✓ mcpfy.config.ts

➜ mcpfy dev

ready on :4141 · 6 tools registered

tools/chart-sales.ts

export const chartSales = tool({
  name: "chart_sales",
  input: z.object({ period: z.string() }),
});

2iterate

Take it for a spin before your users do.

Cloud Inspector connects your server to real MCP clients right in the browser. Call a tool, watch the raw JSON-RPC go by, and switch between models.

Cloud Inspector

No install, just open a tab.

Open Inspector

Model swap

Run the same call across GPT, Claude, and Gemini.

Automatic evals

Regression checks on every model and client.

Try Evals

Cloud Inspectorsession #4f2a

San Francisco, CA

72°F

Partly cloudy · H 74° L 58°

now72°

2pm74°

3pm73°

4pm70°

5pm66°

6pm63°

→ tools/call

{
  "name": "display_weather",
  "arguments": { "city": "San Francisco" }
}

← result20041ms

{ "content": [{ "type": "resource" }] }

GPT-5.2

18/18

Claude

18/18

Gemini

17/18

3monitor

Know what's happening out there.

Built-in analytics, session replay, and observability show you exactly how people are using your MCP app, so a regression reaches you before a user has to report it.

Analytics

Traffic, tool calls, and latency in one view.

Session tracking

Play back any conversation start to finish.

Observability

Traces and error alerts the moment something breaks.

Tool calls · 24h24,732

last 24h ↑ 12.4%

By country34 regions

Tracesp50 42ms

tools/call display_weatherclaude142ms
resources/listchatgpt18ms
tools/call search_docscursor842ms
prompts/listcopilot12ms
tools/call fetch_urlclaude503

p50

38ms

p95

112ms

errors

0.2%

Developers

What people say

Priya Raman@priyabuilds

Been meaning to try Mcpfy. Started a ChatGPT app today and wasn't sure it was the right use case. Turns out it very much is.

Marcus Oyelaran@moyelaran

Easily the smoothest Python path for building an agent-facing MCP server. I can iterate from the CLI and test against a real agent in seconds. Most tooling assumes you're only ever in a chat client.

Ines Ferreira@inesferreira

Widget support landing on the open spec matters: the whole community benefits when compatibility is the default instead of an afterthought.

Philipp Braun@pbraun

Great to see Mcpfy shipping in the open. We're running a hack night with them next week to put it through its paces 😉

Victor Tran@vtran

Sharp execution. They've stripped deployment and aggregation down to one endpoint. Data teams and internal dashboards could slot this in and cut weeks off agent builds.

Priya Raman@priyabuilds

Been meaning to try Mcpfy. Started a ChatGPT app today and wasn't sure it was the right use case. Turns out it very much is.

Marcus Oyelaran@moyelaran

Easily the smoothest Python path for building an agent-facing MCP server. I can iterate from the CLI and test against a real agent in seconds. Most tooling assumes you're only ever in a chat client.

Ines Ferreira@inesferreira

Widget support landing on the open spec matters: the whole community benefits when compatibility is the default instead of an afterthought.

Philipp Braun@pbraun

Great to see Mcpfy shipping in the open. We're running a hack night with them next week to put it through its paces 😉

Victor Tran@vtran

Sharp execution. They've stripped deployment and aggregation down to one endpoint. Data teams and internal dashboards could slot this in and cut weeks off agent builds.

Priya Raman@priyabuilds

Been meaning to try Mcpfy. Started a ChatGPT app today and wasn't sure it was the right use case. Turns out it very much is.

Marcus Oyelaran@moyelaran

Easily the smoothest Python path for building an agent-facing MCP server. I can iterate from the CLI and test against a real agent in seconds. Most tooling assumes you're only ever in a chat client.

Ines Ferreira@inesferreira

Widget support landing on the open spec matters: the whole community benefits when compatibility is the default instead of an afterthought.

Philipp Braun@pbraun

Great to see Mcpfy shipping in the open. We're running a hack night with them next week to put it through its paces 😉

Victor Tran@vtran

Sharp execution. They've stripped deployment and aggregation down to one endpoint. Data teams and internal dashboards could slot this in and cut weeks off agent builds.

Tomas Lindqvist@tlindqvist

Solid addition to the MCP client ecosystem. The Inspector alone saved me an afternoon.

Elena Sokolova@esokolova

The open way to connect any model to any MCP server and build agents with real tool access, without getting locked into one vendor's client.

Darren Cole@dcole

Mcpfy is really onto something here.

Jeffrey Nwosu@jeffnwosu

MCP evals are finally a first-class thing and not a pile of bash scripts I maintain alone 👀

Aisha Mahmood@aishamah

Just found this: an open library that connects any LLM to MCP tools for custom agents, with browsing, search and file access working out of the box.

Tomas Lindqvist@tlindqvist

Solid addition to the MCP client ecosystem. The Inspector alone saved me an afternoon.

Elena Sokolova@esokolova

The open way to connect any model to any MCP server and build agents with real tool access, without getting locked into one vendor's client.

Darren Cole@dcole

Mcpfy is really onto something here.

Jeffrey Nwosu@jeffnwosu

MCP evals are finally a first-class thing and not a pile of bash scripts I maintain alone 👀

Aisha Mahmood@aishamah

Just found this: an open library that connects any LLM to MCP tools for custom agents, with browsing, search and file access working out of the box.

Tomas Lindqvist@tlindqvist

Solid addition to the MCP client ecosystem. The Inspector alone saved me an afternoon.

Elena Sokolova@esokolova

The open way to connect any model to any MCP server and build agents with real tool access, without getting locked into one vendor's client.

Darren Cole@dcole

Mcpfy is really onto something here.

Jeffrey Nwosu@jeffnwosu

MCP evals are finally a first-class thing and not a pile of bash scripts I maintain alone 👀

Aisha Mahmood@aishamah

Just found this: an open library that connects any LLM to MCP tools for custom agents, with browsing, search and file access working out of the box.

Dana Whitfield@danawhit

Infra isn't just plumbing, it's distribution. Whoever owns the rails agents use to talk to the outside world ends up with real leverage.

Reuben Adeyemi@reubendev

Shipped our internal MCP server to production before standup finished. git push, done. 🚀

Hana Kobayashi@hanakb

Open-source dev tools and infrastructure for MCP, helping teams build and deploy custom agents fast. Exactly the layer that was missing.

Clara Bennett@clarabee

Launched! Open infrastructure for MCP agents: spin up and aggregate servers behind a single endpoint with zero friction.

Pietro Zullo@pzullo

Code mode in the client is a great call. Define which servers the agent can use, flip it on, and the agent discovers tools and executes against them as plain modules.

Dana Whitfield@danawhit

Infra isn't just plumbing, it's distribution. Whoever owns the rails agents use to talk to the outside world ends up with real leverage.

Reuben Adeyemi@reubendev

Shipped our internal MCP server to production before standup finished. git push, done. 🚀

Hana Kobayashi@hanakb

Open-source dev tools and infrastructure for MCP, helping teams build and deploy custom agents fast. Exactly the layer that was missing.

Clara Bennett@clarabee

Launched! Open infrastructure for MCP agents: spin up and aggregate servers behind a single endpoint with zero friction.

Pietro Zullo@pzullo

Code mode in the client is a great call. Define which servers the agent can use, flip it on, and the agent discovers tools and executes against them as plain modules.

Dana Whitfield@danawhit

Infra isn't just plumbing, it's distribution. Whoever owns the rails agents use to talk to the outside world ends up with real leverage.

Reuben Adeyemi@reubendev

Shipped our internal MCP server to production before standup finished. git push, done. 🚀

Hana Kobayashi@hanakb

Open-source dev tools and infrastructure for MCP, helping teams build and deploy custom agents fast. Exactly the layer that was missing.

Clara Bennett@clarabee

Launched! Open infrastructure for MCP agents: spin up and aggregate servers behind a single endpoint with zero friction.

Pietro Zullo@pzullo

Code mode in the client is a great call. Define which servers the agent can use, flip it on, and the agent discovers tools and executes against them as plain modules.

FAQ

Answers before you ask

Everything about building, deploying, testing and publishing.

Still stuck? Talk to us

Three ways in

Ship your first MCP server today

Start from the code you already have, an API spec, or a template.

Start deploying Book a call

From a Git repository

Point us at a repo. Every push builds, checks and ships itself.

mcpfy/chart-serverImport

mcpfy/postgres-mcpImport

acme/internal-toolsImport

Connect GitHub

From API Spec

Point us at an OpenAPI spec and get typed tools generated automatically.

openapi.yamlImport

listUsersgetInvoicecreateOrder

Import a spec

From a template

Start on a known-good scaffold: charts, diagrams, databases.

Chart Builder

Diagram Builder

Browse templates

Build and deploy MCP servers and apps

One deploy. Works with every client.

In the ChatGPT Apps Store. In Claude Connectors. From one build.

Why teams pick Mcpfy

From first commit to production

Start from an SDK, a skill, or an API spec.

Take it for a spin before your users do.

Know what's happening out there.

What people say

Answers before you ask

What is Mcpfy?

How fast can I deploy an MCP server on Mcpfy?

How do I test an MCP server across different AI clients?

What is the MCP Inspector?

What analytics do I get for my MCP server?

How does authentication work for MCP servers on Mcpfy?

Do pull requests get preview deployments?

Is the Mcpfy SDK open source?

Can I use Mcpfy without hosting my server there?

Ship your first MCP server today

Build and deploy MCP
servers and apps

In the ChatGPT Apps Store.
In Claude Connectors. From one build.