chatgpt-api/docs/publishing/guides/existing-mcp-server.mdx

358 wiersze
8.7 KiB
Markdown

---
title: Existing MCP Server
description: This guide shows how to publish an existing MCP server to Agentic.
---
<Info>
**Prerequisite**: Please install [Node.js](https://nodejs.org) before
proceeding.
</Info>
## 1. Install the Agentic CLI
<CodeGroup>
```bash npm
npm i -g @agentic/platform-cli
```
```bash pnpm
pnpm i -g @agentic/platform-cli
```
```bash bun
bun install -g @agentic/platform-cli
```
```bash yarn
yarn global add @agentic/platform-cli
```
</CodeGroup>
## 2. Log in or create an account
<Tabs>
<Tab title="GitHub">
The `agentic` CLI defaults to using GitHub for authentication.
```bash
agentic login
```
</Tab>
<Tab title="Username & password">
```bash
agentic login -e <email> -p <password>
# or
agentic signup -e <email> -p <password> -u <username>
```
</Tab>
</Tabs>
## 3. Add an Agentic config to your project
<Info>
Make sure your remote MCP server is deployed to a publicly accessible `https`
URL and that the URL supports the Streamable HTTP transport.
</Info>
<Tabs>
<Tab title="TypeScript">
First, install the `@agentic/platform` package as a dev dependency.
<CodeGroup>
```bash npm
npm i -save-dev @agentic/platform
```
```bash pnpm
pnpm add -D @agentic/platform
```
```bash bun
bun add -d @agentic/platform
```
</CodeGroup>
This package exports a `defineConfig` function which makes your config fully-typed and adds nice autocomplete.
Now, create an `agentic.config.ts` file in the root of your project's source.
```ts agentic.config.ts
import { defineConfig } from '@agentic/platform'
export default defineConfig({
name: '<Your Project Name>',
description: '<A brief description of your project>',
origin: {
type: 'mcp',
url: '<Your Remote MCP Server URL>'
}
})
```
</Tab>
<Tab title="JSON">
Create an `agentic.config.json` file in the root of your project's source.
```json agentic.config.json
{
"name": "<Your Project Name>",
"description": "<Your Project Description>",
"origin": {
"type": "mcp",
"url": "<Your Remote MCP Server URL>"
}
}
```
</Tab>
</Tabs>
## 4. Deploy your project
From the directory where your `agentic.config.ts` or `agentic.config.json` file is located, run:
```bash
agentic deploy
```
Every time you make a change to your project, you can run `agentic deploy` which will create a new immutable preview deployment. These deployments will not affect any published products you may have until you publish them by running `agentic publish`.
<Info>
You'll soon be able to configure a GitHub repository to automatically deploy
your project on changes. Please [let me know](/contact) if you'd like me to
prioritize this feature.
</Info>
<Note>
The returned deployment will not have any information about the origin server,
because your origin server is considered hidden once deployed to Agentic's MCP
gateway.
</Note>
<Expandable title="example output">
```json
{
"id": "depl_m2yl7dpdpc5xk8b3cwvuzkg3",
"createdAt": "2025-06-27 18:14:34.641308+00",
"updatedAt": "2025-06-27 18:14:34.641308+00",
"identifier": "@dev/search@b57dc301",
"hash": "b57dc301",
"published": false,
"description": "Official Google Search tool. Useful for finding up-to-date news and information about any topic.",
"readme": "",
"userId": "user_bhlpuiioipxilpuq7xaoh1ae",
"projectId": "proj_rxs9jorlwolc3seq8enqgrgc",
"tools": [
{
"name": "search",
"description": "Uses Google Search to return the most relevant web pages for a given query. Useful for finding up-to-date news and information about any topic.",
"inputSchema": {
"type": "object",
"properties": {
"num": {
"type": "integer",
"default": 5,
"description": "Number of results to return"
},
"type": {
"enum": [
"search",
"images",
"videos",
"places",
"news",
"shopping"
],
"type": "string",
"default": "search",
"description": "Type of Google search to perform"
},
"query": {
"type": "string",
"description": "Search query"
}
},
"required": ["query"],
"additionalProperties": false,
"$schema": "http://json-schema.org/draft-07/schema#"
},
"outputSchema": {
"type": "object",
"properties": {
"news": {},
"images": {},
"places": {},
"videos": {},
"results": {},
"shopping": {},
"answerBox": {},
"knowledgeGraph": {}
},
"additionalProperties": false,
"$schema": "http://json-schema.org/draft-07/schema#"
}
}
],
"toolConfigs": [
{
"name": "search",
"cacheControl": "public, max-age=60, s-maxage=60 stale-while-revalidate=10"
}
],
"pricingPlans": [
{
"name": "Free",
"slug": "free",
"rateLimit": {
"interval": 86400,
"limit": 10,
"mode": "approximate",
"enabled": true
},
"lineItems": [
{
"slug": "requests",
"usageType": "metered",
"billingScheme": "per_unit",
"unitAmount": 0
}
]
},
{
"name": "Standard",
"slug": "standard",
"interval": "month",
"rateLimit": {
"interval": 1,
"limit": 100,
"mode": "approximate",
"enabled": true
},
"lineItems": [
{
"slug": "base",
"usageType": "licensed",
"amount": 1000
},
{
"slug": "requests",
"usageType": "metered",
"billingScheme": "tiered",
"tiersMode": "volume",
"tiers": [
{
"unitAmount": 0,
"upTo": 1000
},
{
"unitAmount": 0.01,
"upTo": 50000
},
{
"unitAmount": 0.008,
"upTo": 500000
},
{
"unitAmount": 0.006,
"upTo": 2500000
},
{
"unitAmount": 0.005,
"upTo": "inf"
}
]
}
]
}
],
"pricingIntervals": ["month"],
"defaultRateLimit": {
"interval": 60,
"limit": 1000,
"mode": "approximate",
"enabled": true
}
}
```
</Expandable>
## 5. Test your deployment
The easiest way to test your deployment is to visit it in your Agentic dashboard: `https://agentic.so/app/projects/<your-deployment-identifier>`.
This page will show all the tools available on your deployment and includes a GUI for how to call them with various MCP clients, TS LLM SDKs, Python LLM SDKs, and raw HTTP.
<Frame caption='Example of calling an Agentic tool'>
<img
src='/media/example-usage.png'
alt='Example of calling an Agentic tool'
/>
</Frame>
<Expandable title="Example of calling a tool via HTTP">
This example uses the [@agentic/search](https://agentic.so/marketplace/projects/@agentic/search) project's `search` tool. You'll need to replace the project identifier, tool name, and tool arguments with your own, but otherwise, calling your deployment's tools should be pretty straightforward.
<Tabs>
<Tab title="cURL">
```bash
curl -X POST -H "Content-Type: application/json" -d '{ "query": "example google search" }' https://gateway.agentic.com/mcp/search/search
```
</Tab>
<Tab title="HTTPie">
```bash
http -j https://gateway.agentic.com/mcp/search/search query='example google search'
```
</Tab>
</Tabs>
</Expandable>
## 6. Publish your deployment
Publishing your deployment will make it publicly available to all Agentic users. This will also enable other users to subscribe to your product using Stripe subscriptions.
```bash
agentic publish
```
The CLI will prompt you to confirm a `semver` version.
Now, your project will be available at `https://agentic.so/marketplace/projects/<your-project-identifier>`.
**Upon publishing, your project will be a live, publicly available product**.
You can share your product's public URL with customers, and they'll be able to subscribe to your product via Stripe. You can visit your [dashboard](https://agentic.so/app) to track customer usage and revenue.
<Note>Congrats, you now have a live MCP product! 🎉</Note>
## 7. (Optional) Submit your product to the public Agentic Marketplace
<Info>
**Your published product will be live and publicly accessible, but it will not be discoverable on the Agentic Marketplace's main page or search by default.**
I made this decision during the current beta in order to keep the Agentic Marketplace as high quality and curated as possible.
If you'd like to submit your product to the Agentic's public MCP Marketplace, please
[get in touch](/contact).
</Info>