from line 1
11# Model Context Protocol (MCP)
22
3> Understanding the open standard powering Claude's connectors
3> Get oriented to the Model Context Protocol, the open standard behind Claude's connectors: local and remote servers, tools, resources, and prompts.
44
5The Model Context Protocol (MCP) is an open standard created by Anthropic for AI applications to connect with tools and data sources.
5The Model Context Protocol (MCP) is an open standard created by Anthropic for AI applications to connect with tools and data sources. When you build an MCP server and someone adds it to Claude, they see it as a [connector](/docs/connectors/getting-started).
66
7## What is MCP?
7This page is a short orientation for developers who are new to MCP. The [MCP documentation](https://modelcontextprotocol.io/docs) is the source of truth for building MCP servers, and [Build an MCP server for Claude](/docs/connectors/building/index) lists what Claude's client supports once you're ready to build.
88
9MCP provides a standardized way for AI assistants like Claude to:
9## Understand what MCP provides
1010
11MCP gives AI assistants like Claude a standardized way to do the following:
12
1113* Connect to external tools and services
1214* Access data from various sources
1315* Perform actions on behalf of users
1416* Maintain security and user control
1517
16## How MCP works
18## Understand how MCP servers work
1719
18### Local vs remote servers
20An MCP server runs either on the user's device or on the internet, and exposes tools, resources, and prompts to Claude.
1921
20| Type | Description | Use Case |
21| -------------- | ---------------------- | --------------------------------- |
22| **Local MCP** | Runs on your device | Desktop integrations, local tools |
23| **Remote MCP** | Hosted on the internet | Web services, cloud applications |
22### Local and remote servers
2423
25### Key components
24Where the server runs determines which integrations it suits.
2625
27* **Tools**: Actions Claude can perform (search, create, modify)
28* **Resources**: Data Claude can access (files, documents, records)
29* **Prompts**: Predefined interactions for specific tasks
26| Type | Description | Use case |
27| ---------- | ------------------------- | --------------------------------- |
28| Local MCP | Runs on the user's device | Desktop integrations, local tools |
29| Remote MCP | Hosted on the internet | Web services, cloud applications |
3030
31### Tools, resources, and prompts
32
33A server exposes its capabilities to Claude as tools, resources, and prompts:
34
35* **Tools**: actions Claude can perform, such as search, create, and modify
36* **Resources**: data Claude can access, such as files, documents, and records
37* **Prompts**: predefined interactions for specific tasks
38
3139## Security model
3240
41Users stay in control of each connector, and every tool declares whether it can change data.
42
3343### User control
3444
35* You authenticate each connector individually
36* Permissions mirror your access on the external service
37* You can disconnect at any time
45Each person who uses your connector keeps these controls:
3846
47* They authenticate each connector individually
48* Their permissions mirror their access on the external service
49* They can disconnect at any time
50
3951### Tool hints
4052
41All MCP tools must declare:
53All MCP tools must declare both of these annotations:
4254
43* [`readOnlyHint`](https://modelcontextprotocol.io/specification/latest/schema#toolannotations-readonlyhint): Tool only reads data
44* [`destructiveHint`](https://modelcontextprotocol.io/specification/latest/schema#toolannotations-destructivehint): Tool can modify or delete data
55* [`readOnlyHint`](https://modelcontextprotocol.io/specification/latest/schema#toolannotations-readonlyhint): the tool only reads data
56* [`destructiveHint`](https://modelcontextprotocol.io/specification/latest/schema#toolannotations-destructivehint): the tool can modify or delete data
4557
46This helps Claude and users understand what actions are possible.
58Claude and users read these hints to understand what actions a tool can take.
4759
48## Building with MCP
60## Build with MCP
4961
50The [MCP documentation](https://modelcontextprotocol.io/docs) is the source of truth for building MCP servers.
62The [MCP documentation](https://modelcontextprotocol.io/docs) is the source of truth for building MCP servers. Start from these resources:
5163
52### For developers
53
5464* Open specification at [modelcontextprotocol.io](https://modelcontextprotocol.io)
55* [TypeScript SDK](https://github.com/modelcontextprotocol/typescript-sdk) and [Python SDK](https://github.com/modelcontextprotocol/python-sdk) available
65* [TypeScript SDK](https://github.com/modelcontextprotocol/typescript-sdk) and [Python SDK](https://github.com/modelcontextprotocol/python-sdk)
5666* Cloudflare hosting support with OAuth
5767
58### Submitting to directory
59
6068Organizations can [submit MCP servers](/docs/connectors/building/submission) to the Connectors Directory for broader availability.
6169
62## Related topics
70## Next steps
6371
64<Columns cols={2}>
65 <Card title="MCP in Claude Code" icon="terminal" href="https://code.claude.com/docs/en/mcp">
66 Connect MCP servers to Claude Code from the command line.
67 </Card>
68
69 <Card title="Submit to Directory" icon="paper-plane" href="/docs/connectors/building/submission">
70 Review requirements and submit your connector.
71 </Card>
72</Columns>
72* [Build an MCP server for Claude](/docs/connectors/building/index): the transports, authentication, protocol features, and limits Claude's client supports
73* [Authentication for connectors](/docs/connectors/building/authentication): OAuth requirements and supported auth types
74* [MCP in Claude Code](https://code.claude.com/docs/en/mcp): connect MCP servers to Claude Code from the command line
75* [Submitting to the Connectors Directory](/docs/connectors/building/submission): review requirements and submit your connector
7376
No line in this hunk matches that.