A clojure-sdk for creating Model Context Protocol servers!

• io.modelcontext/clojure-sdk
  • Usage
    • Deps
    • Templates for Quickstart
    • Examples
      • Building the Examples Jar
      • Calculator: calculator_server
      • Vega-lite: vegalite_server
      • Code Analysis: code_analysis_server
  • Core Components
  • Communication Flow
  • Pending Work
  • Development of the SDK
  • Inspiration
  • License

The calculator_server.clj file and vegalite_server.clj file examples contains a full working code for defining an MCP server.

examples is a deps-new app project, and instructions for compiling and running the various example servers are in the examples/README.md file (also copied below this section)

The deps for clojure-sdk are:

{io.modelcontextprotocol/mcp-clojure-sdk {:git/url "https://github.com/unravel-team/mcp-clojure-sdk.git" :git/sha "039cf220ac6bb3858f71e823016035e257a5380d"}}

For your ease of use, there is also a deps-new template and a Github template. See:

1. mcp-clojure-server-deps-new for a deps-new based template to quickly create new MCP servers.
2. example-cool-mcp-server for a Github template project to quickly create new MCP servers.

$ make clean && make examples-jar 

The examples jar contains the following servers:

1. Calculator: calculator_server
2. Vega-lite: vegalite_server
3. Code Analysis: code_analysis_server

Provides basic arithmetic tools: add, subtract, multiply, divide, power, square-root, average, factorial

Some example commands you can try in Claude Desktop or Inspector:

1. What's the average of [1, 2, 3, 4, 5]?
2. What's the factorial of 15?
3. What's 2 to the power of 1000?
4. What's the square-root of 64?

Remember:

1. Use the full-path to the examples JAR on your system

 "calculator": { "command": "java", "args": [ "-Dclojure.tools.logging.factory=clojure.tools.logging.impl/log4j2-factory", "-Dorg.eclipse.jetty.util.log.class=org.eclipse.jetty.util.log.Slf4jLog", "-Dlog4j2.contextSelector=org.apache.logging.log4j.core.async.AsyncLoggerContextSelector", "-Dlog4j2.configurationFile=log4j2-mcp.xml", "-Dbabashka.json.provider=metosin/jsonista", "-Dlogging.level=INFO", "-cp", "/Users/vedang/mcp-clojure-sdk/examples/target/io.modelcontextprotocol.clojure-sdk/examples-1.2.0.jar", "calculator_server" ] }

npx @modelcontextprotocol/inspector java -Dclojure.tools.logging.factory=clojure.tools.logging.impl/log4j2-factory -Dorg.eclipse.jetty.util.log.class=org.eclipse.jetty.util.log.Slf4jLog -Dlog4j2.contextSelector=org.apache.logging.log4j.core.async.AsyncLoggerContextSelector -Dlog4j2.configurationFile=log4j2-mcp.xml -Dbabashka.json.provider=metosin/jsonista -Dlogging.level=INFO -cp examples/target/io.modelcontextprotocol.clojure-sdk/examples-1.2.0.jar calculator_server

Provides tools for generating Vega-lite charts: save-data, visualize-data.

PRE-REQUISITES: Needs vl-convert CLI to be installed.

Some example commands you can try in Claude Desktop or Inspector:

Here is some example data for you:

[ { "year": 2011, "value": 14.6, "growth_type": "Market Cap Growth" }, { "year": 2011, "value": 11.4, "growth_type": "Revenue Growth" }, { "year": 2011, "value": 26.6, "growth_type": "Net Income Growth" }, { "year": 2012, "value": 40.1, "growth_type": "Market Cap Growth" }, { "year": 2012, "value": 42.7, "growth_type": "Revenue Growth" }, { "year": 2012, "value": 36.9, "growth_type": "Net Income Growth" }, { "year": 2013, "value": 16.9, "growth_type": "Market Cap Growth" }, { "year": 2013, "value": 14.6, "growth_type": "Revenue Growth" }, { "year": 2013, "value": 15.3, "growth_type": "Net Income Growth" }, { "year": 2014, "value": 9.6, "growth_type": "Market Cap Growth" }, { "year": 2014, "value": 7.9, "growth_type": "Revenue Growth" }, { "year": 2014, "value": 10.9, "growth_type": "Net Income Growth" }, { "year": 2015, "value": 5.8, "growth_type": "Market Cap Growth" }, { "year": 2015, "value": 6.7, "growth_type": "Revenue Growth" }, { "year": 2015, "value": 6.2, "growth_type": "Net Income Growth" }, { "year": 2016, "value": -12.4, "growth_type": "Market Cap Growth" }, { "year": 2016, "value": -3.9, "growth_type": "Revenue Growth" }, { "year": 2016, "value": -32.2, "growth_type": "Net Income Growth" }, { "year": 2017, "value": 25.3, "growth_type": "Market Cap Growth" }, { "year": 2017, "value": 5.9, "growth_type": "Revenue Growth" }, { "year": 2017, "value": 43.9, "growth_type": "Net Income Growth" } ]

Visualize this data for me using vega-lite.

Remember:

1. Replace the full-path to the examples JAR with the correct path on your system
2. Specify the full-path to vl-convert on your system

 "vegalite": { "command": "java", "args": [ "-Dclojure.tools.logging.factory=clojure.tools.logging.impl/log4j2-factory", "-Dorg.eclipse.jetty.util.log.class=org.eclipse.jetty.util.log.Slf4jLog", "-Dlog4j2.contextSelector=org.apache.logging.log4j.core.async.AsyncLoggerContextSelector", "-Dlog4j2.configurationFile=log4j2-mcp.xml", "-Dbabashka.json.provider=metosin/jsonista", "-Dlogging.level=INFO", "-Dmcp.vegalite.vl_convert_executable=/Users/vedang/.cargo/bin/vl-convert", "-cp", "/Users/vedang/mcp-clojure-sdk/examples/target/io.modelcontextprotocol.clojure-sdk/examples-1.2.0.jar", "vegalite_server" ] }

Remember to use the full-path to the examples JAR on your system, or execute this command from the mcp-clojure-sdk repo.

npx @modelcontextprotocol/inspector java -Dclojure.tools.logging.factory=clojure.tools.logging.impl/log4j2-factory -Dorg.eclipse.jetty.util.log.class=org.eclipse.jetty.util.log.Slf4jLog -Dlog4j2.contextSelector=org.apache.logging.log4j.core.async.AsyncLoggerContextSelector -Dlog4j2.configurationFile=log4j2-mcp.xml -Dbabashka.json.provider=metosin/jsonista -Dlogging.level=INFO -Dmcp.vegalite.vl_convert_executable=/Users/vedang/.cargo/bin/vl-convert -cp examples/target/io.modelcontextprotocol.clojure-sdk/examples-1.2.0.jar vegalite_server

This server is an example of a server which provides prompts and not tools. The following prompts are available: analyse-code and poem-about-code.

You can try the prompts out in Claude Desktop or Inspector. While these prompts are very basic, this is a good way to see how you could expose powerful prompts through this technique.

Remember:

1. Replace the full-path to the examples JAR with the correct path on your system

 "code-anaylsis": { "command": "java", "args": [ "-Dclojure.tools.logging.factory=clojure.tools.logging.impl/log4j2-factory", "-Dorg.eclipse.jetty.util.log.class=org.eclipse.jetty.util.log.Slf4jLog", "-Dlog4j2.contextSelector=org.apache.logging.log4j.core.async.AsyncLoggerContextSelector", "-Dlog4j2.configurationFile=log4j2-mcp.xml", "-Dbabashka.json.provider=metosin/jsonista", "-Dlogging.level=INFO", "-cp", "/Users/vedang/mcp-clojure-sdk/examples/target/io.modelcontextprotocol.clojure-sdk/examples-1.2.0.jar", "code_analysis_server" ] }

(Remember to use the full-path to the examples JAR on your system, or execute this command from the mcp-clojure-sdk repo)

npx @modelcontextprotocol/inspector java -Dclojure.tools.logging.factory=clojure.tools.logging.impl/log4j2-factory -Dorg.eclipse.jetty.util.log.class=org.eclipse.jetty.util.log.Slf4jLog -Dlog4j2.contextSelector=org.apache.logging.log4j.core.async.AsyncLoggerContextSelector -Dlog4j2.configurationFile=log4j2-mcp.xml -Dbabashka.json.provider=metosin/jsonista -Dlogging.level=INFO -cp examples/target/io.modelcontextprotocol.clojure-sdk/examples-1.2.0.jar code_analysis_server

1. Server Implementation: The core server functionality is implemented in server.clj, which handles request/response cycles for various MCP methods.

2. Transport Layer: The SDK implements a STDIO transport in stdio_server.clj using io_chan.clj to convert between IO streams and core.async channels.

3. Error Handling: Custom error handling is defined in mcp/errors.clj.

4. Protocol Specifications: All protocol specifications are defined in specs.clj, which provides validation for requests, responses, and server components.

The sequence diagram shows the typical lifecycle of an MCP client-server interaction:

1. Initialization Phase:

   • The client connects and sends an initialize request
   • The server responds with its capabilities
   • The client confirms with an initialized notification

2. Discovery Phase:

   • The client discovers available tools, resources, and prompts using methods tools/list, resources/list and prompts/list
   • These are registered in the server during context creation

3. Tool Interaction:

   • The client can call tools with arguments
   • The server routes these to the appropriate handler function
   • Results are returned to the client

4. Resource Interaction:

   • The client can read resources by URI
   • The server retrieves the resource content

5. Prompt Interaction:

   • The client can request predefined prompts
   • The server returns the appropriate messages

6. Optional Features:

   • Resource subscription for updates
   • Health checks via ping/pong

sequenceDiagram participant Client participant MCPServer participant Tool participant Resource participant Prompt Note over Client,MCPServer: Initialization Phase Client->>+MCPServer: initialize MCPServer-->>-Client: initialize response (capabilities) Client->>MCPServer: notifications/initialized Note over Client,MCPServer: Discovery Phase Client->>+MCPServer: tools/list MCPServer-->>-Client: List of available tools Client->>+MCPServer: resources/list MCPServer-->>-Client: List of available resources Client->>+MCPServer: prompts/list MCPServer-->>-Client: List of available prompts Note over Client,MCPServer: Tool Interaction Client->>+MCPServer: tools/call (name, arguments) MCPServer->>+Tool: handler(arguments) Tool-->>-MCPServer: result MCPServer-->>-Client: Tool response Note over Client,MCPServer: Resource Interaction Client->>+MCPServer: resources/read (uri) MCPServer->>+Resource: handler(uri) Resource-->>-MCPServer: contents MCPServer-->>-Client: Resource contents Note over Client,MCPServer: Prompt Interaction Client->>+MCPServer: prompts/get (name, arguments) MCPServer->>+Prompt: handler(arguments) Prompt-->>-MCPServer: messages MCPServer-->>-Client: Prompt messages Note over Client,MCPServer: Optional Subscription Client->>+MCPServer: resources/subscribe (uri) MCPServer-->>-Client: Empty response MCPServer-->>Client: notifications/resources/updated Note over Client,MCPServer: Health Check Client->>+MCPServer: ping MCPServer-->>-Client: pong 

You can help dear reader! Head over to the todo.org file to see the list of pending changes, arranged roughly in the order I plan to tackle them.

The clojure-sdk is a standard deps-new project, so you should expect all the deps-new commands to work as expected. Even so:

Run the project's tests:

$ make test ## or clojure -T:build test 

Run the project's CI pipeline and build a JAR:

$ make build ## or clojure -T:build ci 

This will produce an updated pom.xml file with synchronized dependencies inside the META-INF directory inside target/classes and the JAR in target. You can update the version (and SCM tag) information in generated pom.xml by updating build.clj.

Install it locally:

$ make install ## or clojure -T:build install 

Deploy it to Clojars -- needs CLOJARS_USERNAME and CLOJARS_PASSWORD environment variables (requires the ci task be run first):

$ make deploy ## or clojure -T:build deploy 

Your library will be deployed to io.modelcontext/clojure-sdk on clojars.org by default.

This SDK is built on top of lsp4clj, which solves the hard part of handling all the edge-cases of a JSON-RPC based server. I built this layer by hand and discovered all the edge-cases before realising that lsp4clj was the smarter approach. The code is super well written and easy to modify for my requirements.

Copyright © 2025 Unravel.tech

Distributed under the MIT License