mcp-server: Library for building Model Context Protocol (MCP) servers

[ bsd3, json-rpc, library, mcp, network, program, server, service ] [ Propose Tags ] [ Report a vulnerability ]

A fully featured library for building MCP (Model Context Protocol) servers. Supports both low-level fine-grained handling and high-level derived interfaces for prompts, resources, and tools. Includes JSON-RPC transport and stdin/stdout communication.


[Skip to Readme]

Downloads

Maintainer's Corner

Package maintainers

For package maintainers and hackage trustees

Candidates

Versions [RSS] 0.1.0.0, 0.1.0.1, 0.1.0.2, 0.1.0.3, 0.1.0.4, 0.1.0.5, 0.1.0.6, 0.1.0.7, 0.1.0.8, 0.1.0.9, 0.1.0.10, 0.1.0.11, 0.1.0.12, 0.1.0.13
Change log CHANGELOG.md
Dependencies aeson (>=2.0 && <3.0), base (>=4.20.0 && <4.22), bytestring (>=0.10 && <0.13), containers (>=0.6 && <0.8), http-types (>=0.12 && <1.0), mcp-server, network-uri (>=2.6 && <2.8), template-haskell (>=2.16 && <2.24), text (>=1.2 && <3.0), vector (>=0.12 && <1.0), wai (>=3.2 && <4.0), warp (>=3.3 && <4.0) [details]
License BSD-3-Clause
Copyright 2025 Tom Wells
Author Tom Wells
Maintainer drshade@gmail.com
Category Network, Server, Service, MCP, JSON-RPC
Home page https://github.com/drshade/haskell-mcp-server
Bug tracker https://github.com/drshade/haskell-mcp-server/issues
Source repo head: git clone https://github.com/drshade/haskell-mcp-server.git
Uploaded by drshade at 2025-06-17T14:19:20Z
Distributions
Executables http-simple-example, complete-example, simple-example, haskell-mcp-server
Downloads 40 total (40 in the last 30 days)
Rating 2.0 (votes: 1) [estimated by Bayesian average]
Your Rating
  • λ
  • λ
  • λ
Status Docs uploaded by user
Build status unknown [no reports yet]

Readme for mcp-server-0.1.0.13

[back to package description]

mcp-server

A fully-featured Haskell library for building Model Context Protocol (MCP) servers.

Features

  • Complete MCP Implementation: Supports MCP 2025-03-26 specification (with backward compatibility for 2024-11-05)
  • Type-Safe API: Leverage Haskell's type system for robust MCP servers
  • Multiple Abstractions: Both low-level fine-grained control and high-level derived interfaces
  • Template Haskell Support: Automatic handler derivation from data types
  • Multiple Transports: STDIO and HTTP Streaming transport (MCP 2025-03-26 Streamable HTTP)

Supported MCP Features

  • Prompts: User-controlled prompt templates with arguments
  • Resources: Application-controlled readable resources
  • Tools: Model-controlled callable functions
  • Initialization Flow: Complete protocol lifecycle with version negotiation
  • Error Handling: Comprehensive error types and JSON-RPC error responses

Quick Start

Add the library mcp-server to your cabal file:

build-depends:
  mcp-server

Create a simple module, such as this example below:

{-# LANGUAGE OverloadedStrings #-}
{-# LANGUAGE TemplateHaskell #-}

import MCP.Server
import MCP.Server.Derive

-- Define your data types
data MyPrompt = Recipe { idea :: Text } | Shopping { items :: Text }
data MyResource = Menu | Specials  
data MyTool = Search { query :: Text } | Order { item :: Text }

-- Implement handlers
handlePrompt :: MyPrompt -> IO Content
handlePrompt (Recipe idea) = pure $ ContentText $ "Recipe for " <> idea
handlePrompt (Shopping items) = pure $ ContentText $ "Shopping list: " <> items

handleResource :: MyResource -> IO Content  
handleResource Menu = pure $ ContentText "Today's menu..."
handleResource Specials = pure $ ContentText "Daily specials..."

handleTool :: MyTool -> IO Content
handleTool (Search query) = pure $ ContentText $ "Search results for " <> query
handleTool (Order item) = pure $ ContentText $ "Ordered " <> item

-- Derive handlers automatically
main :: IO ()
main = runMcpServerStdio serverInfo handlers
  where
    serverInfo = McpServerInfo
      { serverName = "My MCP Server"
      , serverVersion = "1.0.0" 
      , serverInstructions = "A sample MCP server"
      }
    handlers = McpServerHandlers
      { prompts = Just $(derivePromptHandler ''MyPrompt 'handlePrompt)
      , resources = Just $(deriveResourceHandler ''MyResource 'handleResource)  
      , tools = Just $(deriveToolHandler ''MyTool 'handleTool)
      }

Nestable data types

You can also nest your data types, but MUST always end in a record with named fields:

-- Using fields, but prefixing with the Type name, because maybe the user wants to use Lenses
data GetValueParams = GetValueParams { _gvpKey :: Text }
data SetValueParams = SetValueParams { _svpKey :: Text, _svpValue :: Text }

data SimpleTool
    = GetValue GetValueParams
    | SetValue SetValueParams
    deriving (Show, Eq)

Which is probably nicer for using things like Lenses etc. However we do not support positional (and unnamed) parameters such as:

-- Positional arguments
data SimpleTool
    = GetValue Int
    | SetValue Int Text
    deriving (Show, Eq)

Because we don't want to be generating names when returning details in MCP definitions.

Custom Descriptions

You can provide custom descriptions for constructors and fields using the *WithDescription variants:

-- Define descriptions for constructors and fields
descriptions :: [(String, String)]
descriptions = 
  [ ("Recipe", "Generate a recipe for a specific dish")     -- Constructor description
  , ("Search", "Search our menu database")                  -- Constructor description
  , ("idea", "The dish you want a recipe for")              -- Field description
  , ("query", "Search terms to find menu items")            -- Field description
  ]

-- Use in derivation
handlers = McpServerHandlers
  { prompts = Just $(derivePromptHandlerWithDescription ''MyPrompt 'handlePrompt descriptions)
  , tools = Just $(deriveToolHandlerWithDescription ''MyTool 'handleTool descriptions)
  , resources = Just $(deriveResourceHandlerWithDescription ''MyResource 'handleResource descriptions)
  }

Manual Handler Implementation

For fine-grained control, implement handlers manually:

import MCP.Server

-- Manual handler implementation
promptListHandler :: IO [PromptDefinition]
promptGetHandler :: PromptName -> [(ArgumentName, ArgumentValue)] -> IO (Either Error Content)
-- ... implement your custom logic

main :: IO ()
main = runMcpServerStdio serverInfo handlers
  where
    handlers = McpServerHandlers
      { prompts = Just (promptListHandler, promptGetHandler)
      , resources = Nothing  -- Not supported
      , tools = Nothing      -- Not supported  
      }

HTTP Transport (NEW!)

The library now supports MCP 2025-03-26 Streamable HTTP transport:

import MCP.Server.Transport.Http

-- Simple HTTP server (localhost:3000/mcp)
main = runMcpServerHttp serverInfo handlers

-- Custom configuration
main = runMcpServerHttpWithConfig customConfig serverInfo handlers
  where
    customConfig = HttpConfig
      { httpPort = 8080
      , httpHost = "0.0.0.0"
      , httpEndpoint = "/api/mcp"
      , httpVerbose = True  -- Enable detailed logging
      }

Features:

  • JSON-RPC batching support
  • CORS enabled for web clients
  • GET /mcp for server discovery
  • POST /mcp for JSON-RPC messages
  • Full MCP 2025-03-26 compliance

Examples

The library includes several examples:

  • examples/Simple/: Basic key-value store using Template Haskell derivation (STDIO)
  • examples/Complete/: Full-featured example with prompts, resources, and tools (STDIO)
  • examples/HttpSimple/: HTTP version of the simple key-value store

Docker Usage

I like to build and publish my MCP servers to Docker - which means that it's much easier to configure assistants such as Claude Desktop to run them.

# Build the image
docker build -t haskell-mcp-server .

# Run different examples
docker run -i --entrypoint="/usr/local/bin/haskell-mcp-server" haskell-mcp-server

And then configure Claude by editing claude_desktop_config.json:

{
    "mcpServers": {
       "haskell-mcp-server-example": {
            "command": "docker",
            "args": [
                "run",
                "-i",
                "--entrypoint=/usr/local/bin/haskell-mcp-server",
                "haskell-mcp-server"
            ]
        }
    }
}

Documentation

Contributing

Contributions are welcome! Please see the issue tracker for open issues and feature requests.

Disclaimer - AI Assistance

I am not sure whether there is any stigma associated with this but Claude helped me write a lot of this library. I started with a very specific specification of what I wanted to achieve and worked shoulder-to-shoulder with Claude to implement and refactor the library until I was happy with it. A few of the features such as the Derive functions are a little out of my comfort zone to have manually written, so I appreciated having an expert guide me here - however I do suspect that this implementation may be sub-par and I do intend to refactor and rewrite large pieces of this through regular maintenance.

License

BSD-3-Clause