New

Stainless Docs Platform - Early Access

New

Stainless Docs Platform

Production-ready MCP servers optimized for agentic coding and context limits

Production-ready
MCP servers optimized for agentic coding and context limits

Production-ready MCP servers optimized for agentic coding and context limits

Deterministically generate robust, optimized, and production-ready MCP servers directly from your OpenAPI spec. Handle complex schemas, client quirks, and API maintenance automatically.

Deterministically generate robust, optimized, and production-ready MCP servers directly from your OpenAPI spec. Handle complex schemas, client quirks, and API maintenance automatically.

Deterministically generate robust, optimized, and production-ready MCP servers directly from your OpenAPI spec. Handle complex schemas, client quirks, and API maintenance automatically.

View Documentation

View Documentation

Get started

Get started

Trusted by top API companies

Trusted by top API companies

Trusted by top API companies

Agents

Agents

Agents

Supercharge Agentic Coding

Supercharge Agentic Coding

Supercharge Agentic Coding

Your user’s development teams are already using AI to write code. The critical question is whether those AI assistants can correctly integrate with your API.

Stainless generates MCP servers that act as a dedicated interface for AI coding agents. Empower agent mode in editors like Cursor to intelligently query your API's structure, understand specific endpoints, and write the code to integrate with them. This allows your developers to build integrations at the speed of thought, with the precision of a well-engineered tool.

Your user’s development teams are already using AI to write code. The critical question is whether those AI assistants can correctly integrate with your API.

Stainless generates MCP servers that act as a dedicated interface for AI coding agents. Empower agent mode in editors like Cursor to intelligently query your API's structure, understand specific endpoints, and write the code to integrate with them. This allows your developers to build integrations at the speed of thought, with the precision of a well-engineered tool.

Your user’s development teams are already using AI to write code. The critical question is whether those AI assistants can correctly integrate with your API.

Stainless generates MCP servers that act as a dedicated interface for AI coding agents. Empower agent mode in editors like Cursor to intelligently query your API's structure, understand specific endpoints, and write the code to integrate with them. This allows your developers to build integrations at the speed of thought, with the precision of a well-engineered tool.

“Stainless has been a wonderful partner in helping us generate SDKs for Cloudflare’s API and is now one of the fastest ways to get up and running with MCP servers. Building great MCP servers requires tasteful product thinking, and we’re excited that Stainless is already thinking beyond just mapping REST APIs to MCP tools 1:1.”

Brendan Irvine-Broque

Director of Product, Cloudflare Workers

“Stainless has been a wonderful partner in helping us generate SDKs for Cloudflare’s API and is now one of the fastest ways to get up and running with MCP servers. Building great MCP servers requires tasteful product thinking, and we’re excited that Stainless is already thinking beyond just mapping REST APIs to MCP tools 1:1.”

Brendan Irvine-Broque

Director of Product, Cloudflare Workers

“Stainless has been a wonderful partner in helping us generate SDKs for Cloudflare’s API and is now one of the fastest ways to get up and running with MCP servers. Building great MCP servers requires tasteful product thinking, and we’re excited that Stainless is already thinking beyond just mapping REST APIs to MCP tools 1:1.”

Brendan Irvine-Broque

Director of Product, Cloudflare Workers

Tools

Tools

Tools

Dynamic tool discovery for large APIs

Dynamic tool
discovery for
large APIs

Dynamic tool discovery for large APIs

Instead of overflowing context window limits with hundreds of static tools, dynamic discovery lets LLMs search for exactly what they need with just three meta-tools: list, get, and invoke. Your API stays fully accessible even if you have a thousand endpoints.

Tokens

Tokens

Tokens

Minimal token usage

LLMs can filter API responses on-the-fly with jq and extract only the data needed for each query. This dramatically reduces token consumption without limiting access to your API's full capabilities.

# Configure for Claude AI
npx -y my-org-mcp --client=claude

# Configure for Cursor
npx -y my-org-mcp --client=cursor
# Configure for Claude AI
npx -y my-org-mcp --client=claude

# Configure for Cursor
npx -y my-org-mcp --client=cursor
# Configure for Claude AI
npx -y my-org-mcp --client=claude

# Configure for Cursor
npx -y my-org-mcp --client=cursor

Schemas

Schemas

Schemas

Client-optimized tool schemas

Production-grade
SDKs

Client-optimized tool schemas

Different LLM clients handle JSON schemas provided by MCP servers differently. The --client argument automatically adapts tool schemas to each client's specific limitations, gracefully handling OpenAI's union types, Cursor's 60-character name limit, and more. Ship once, serve everywhere.

"This dramatically reduced development time and ensured a high-quality experience for our team and our MCP users."

Darius Emrani

Founder/CEO

"This dramatically reduced development time and ensured a high-quality experience for our team and our MCP users."

Darius Emrani

Founder/CEO

"This dramatically reduced development time and ensured a high-quality experience for our team and our MCP users."

Darius Emrani

Founder/CEO

Tools

Tools

Tools

Granular tool exposure

Provide power users with fine-grained control. Define which tools are available in your Stainless configuration, then allow users to further scope their context with flags like --tool, --resource, or --operation. You expose your API, then let power users further optimize their context window.

resources:
  messages:
    mcp: true # enable MCP generation for all methods in this resource
    methods: ...
  fine_tunes:
    methods:
      create:
        mcp: true # enable this method for MCP generation
        endpoint: post /v1/fine_tunes
      update:


resources:
  messages:
    mcp: true # enable MCP generation for all methods in this resource
    methods: ...
  fine_tunes:
    methods:
      create:
        mcp: true # enable this method for MCP generation
        endpoint: post /v1/fine_tunes
      update:


resources:
  messages:
    mcp: true # enable MCP generation for all methods in this resource
    methods: ...
  fine_tunes:
    methods:
      create:
        mcp: true # enable this method for MCP generation
        endpoint: post /v1/fine_tunes
      update:


{
  "mcpServers": {
    "my_org_api": {
      "command": "npx",
      "args": ["-y", "--tool=refund"],
      "env": {
        "MY_API_KEY": "your-api-key"


{
  "mcpServers": {
    "my_org_api": {
      "command": "npx",
      "args": ["-y", "--tool=refund"],
      "env": {
        "MY_API_KEY": "your-api-key"


{
  "mcpServers": {
    "my_org_api": {
      "command": "npx",
      "args": ["-y", "--tool=refund"],
      "env": {
        "MY_API_KEY": "your-api-key"

"Thanks to the team at Stainless for their ingenuity in building smart response filtering into the MCP server using jq which keeps large JSON responses from overwhelming model context windows.”

Zeke Sikelianos

Founding Designer - Replicate

"Thanks to the team at Stainless for their ingenuity in building smart response filtering into the MCP server using jq which keeps large JSON responses from overwhelming model context windows.”

Zeke Sikelianos

Founding Designer - Replicate

"Thanks to the team at Stainless for their ingenuity in building smart response filtering into the MCP server using jq which keeps large JSON responses from overwhelming model context windows.”

Zeke Sikelianos

Founding Designer - Replicate

MCP

MCP

MCP

Why Stainless for MCP

Robust by default, extensible by design

With Stainless, you get the best of both worlds. Production-grade capabilities like OAuth, SSE, and remote servers are included by default. And when you want even more control, you can filter which endpoints are exposed and customize server and tool descriptions.

Generate without hiccups

Focus on your core API instead of learning MCP protocols or handling edge cases. Complex OpenAPI features like $refs, unions, and recursive types work automatically with intelligent fallbacks, while streaming protocols and OAuth flows come built-in so you can generate production-ready MCP servers without wrestling with specification complexities.

Polished user experience

Your users benefit from a seamless experience whether they use Claude, OpenAI, Cursor, or other clients. Dynamic discovery surfaces only relevant endpoints, token-efficient responses maximize context windows, and automatic client compatibility ensures that your spec is gracefully modified for any LLM.

Maintenance free

Never worry about spec drift breaking AI integrations again. Automated GitHub workflows detect OpenAPI changes and generate updated MCP servers accordingly.

"The automatic MCP server is an excellent feature and it works very well! I've deleted my hand-rolled MCP repo."

Matthew Blode

Co-founder/CTO

"The automatic MCP server is an excellent feature and it works very well! I've deleted my hand-rolled MCP repo."

Matthew Blode

Co-founder/CTO

"The automatic MCP server is an excellent feature and it works very well! I've deleted my hand-rolled MCP repo."

Matthew Blode

Co-founder/CTO

FAQ

How does Stainless generate MCP servers from my existing API?

Stainless automatically transforms your OpenAPI specification into a production-ready MCP server. Simply add the MCP target to your Stainless config, and we'll generate a complete server with OAuth flows, client compatibility handling, and deployment options. No manual coding of individual tools required.

Can end users choose which tools to use, or do they get everything?

End users have complete flexibility. They can import all tools, use dynamic discovery mode for large APIs (which exposes just 3 meta-tools: list, get schema, invoke), or selectively import specific tools using flags like --tool, --resource, or --operation. You control what's available, they control what they use.

Will my MCP server work with different LLM clients like Claude, OpenAI, and Cursor?

Yes, automatically. End users specify their client with a --client flag (e.g., --client=openai-agents), and the server automatically adapts schemas to handle each client's limitations - splitting unions for OpenAI, handling tool name length limits for Cursor, etc.

How do I deploy and host the MCP server?

You have multiple options: publish as an NPM package for local use, deploy to Cloudflare Workers with one-click deployment and built-in OAuth, or use Docker containers for deployment and distribution. Remote servers automatically handle OAuth flows while local servers use environment variables.

What happens when my API changes? Do I need to manually update the MCP server?

No manual work required. Stainless provides automated GitHub workflows that detect OpenAPI spec changes and regenerate your MCP server automatically with deterministic, merge-conflict-free updates. Your MCP server stays perfectly in sync as your API evolves.

How does this handle large APIs with hundreds of endpoints?

Stainless uses dynamic tool discovery to prevent context window overload. Instead of exposing hundreds of static tools, your API becomes discoverable through 3 meta-tools that let LLMs discover and invoke endpoints on-demand. Plus, LLMs can pre-filter the response they get using jq, which trims down unwanted data in the context window.

How does Stainless generate MCP servers from my existing API?

Stainless automatically transforms your OpenAPI specification into a production-ready MCP server. Simply add the MCP target to your Stainless config, and we'll generate a complete server with OAuth flows, client compatibility handling, and deployment options. No manual coding of individual tools required.

Can end users choose which tools to use, or do they get everything?

End users have complete flexibility. They can import all tools, use dynamic discovery mode for large APIs (which exposes just 3 meta-tools: list, get schema, invoke), or selectively import specific tools using flags like --tool, --resource, or --operation. You control what's available, they control what they use.

Will my MCP server work with different LLM clients like Claude, OpenAI, and Cursor?

Yes, automatically. End users specify their client with a --client flag (e.g., --client=openai-agents), and the server automatically adapts schemas to handle each client's limitations - splitting unions for OpenAI, handling tool name length limits for Cursor, etc.

How do I deploy and host the MCP server?

You have multiple options: publish as an NPM package for local use, deploy to Cloudflare Workers with one-click deployment and built-in OAuth, or use Docker containers for deployment and distribution. Remote servers automatically handle OAuth flows while local servers use environment variables.

What happens when my API changes? Do I need to manually update the MCP server?

No manual work required. Stainless provides automated GitHub workflows that detect OpenAPI spec changes and regenerate your MCP server automatically with deterministic, merge-conflict-free updates. Your MCP server stays perfectly in sync as your API evolves.

How does this handle large APIs with hundreds of endpoints?

Stainless uses dynamic tool discovery to prevent context window overload. Instead of exposing hundreds of static tools, your API becomes discoverable through 3 meta-tools that let LLMs discover and invoke endpoints on-demand. Plus, LLMs can pre-filter the response they get using jq, which trims down unwanted data in the context window.

How does Stainless generate MCP servers from my existing API?

Stainless automatically transforms your OpenAPI specification into a production-ready MCP server. Simply add the MCP target to your Stainless config, and we'll generate a complete server with OAuth flows, client compatibility handling, and deployment options. No manual coding of individual tools required.

Can end users choose which tools to use, or do they get everything?

End users have complete flexibility. They can import all tools, use dynamic discovery mode for large APIs (which exposes just 3 meta-tools: list, get schema, invoke), or selectively import specific tools using flags like --tool, --resource, or --operation. You control what's available, they control what they use.

Will my MCP server work with different LLM clients like Claude, OpenAI, and Cursor?

Yes, automatically. End users specify their client with a --client flag (e.g., --client=openai-agents), and the server automatically adapts schemas to handle each client's limitations - splitting unions for OpenAI, handling tool name length limits for Cursor, etc.

How do I deploy and host the MCP server?

You have multiple options: publish as an NPM package for local use, deploy to Cloudflare Workers with one-click deployment and built-in OAuth, or use Docker containers for deployment and distribution. Remote servers automatically handle OAuth flows while local servers use environment variables.

What happens when my API changes? Do I need to manually update the MCP server?

No manual work required. Stainless provides automated GitHub workflows that detect OpenAPI spec changes and regenerate your MCP server automatically with deterministic, merge-conflict-free updates. Your MCP server stays perfectly in sync as your API evolves.

How does this handle large APIs with hundreds of endpoints?

Stainless uses dynamic tool discovery to prevent context window overload. Instead of exposing hundreds of static tools, your API becomes discoverable through 3 meta-tools that let LLMs discover and invoke endpoints on-demand. Plus, LLMs can pre-filter the response they get using jq, which trims down unwanted data in the context window.

MCP Servers

MCP Servers

MCP Servers

Ready to Generate?

Stop wrestling with MCP protocols. Start shipping.

View documentation

View documentation

View documentation

Get started

Get started

Get started