Skip to content

crumplecup/elicitation

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

80 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Elicitation

Conversational elicitation of strongly-typed Rust values via MCP

Crates.io Documentation License Downloads Rust

elicitation is a Rust library that transforms conversational LLM interactions into strongly-typed Rust values through the Model Context Protocol (MCP). It provides a trait-based system for eliciting primitive types, enums, structs, and nested data structures through natural language interaction.

Features

  • Comprehensive Type Coverage - Elicit virtually any Rust standard library type
  • Trait-Based Design - Simple Elicitation trait for all types
  • Derive Macros - Zero-boilerplate implementation for custom types
  • Type-Safe - Compile-time guarantees with full type inference
  • Composable - Nest types arbitrarily deep without limitations
  • Async-First - Built on tokio with proper Send bounds
  • Four Interaction Paradigms:
    • Select - Choose from finite options (enum pattern)
    • Affirm - Yes/no confirmation (bool pattern)
    • Survey - Multi-field elicitation (struct pattern)
    • Authorize - Permission policies (planned for v0.2.0)
  • MCP Integration - Uses official rmcp (Rust MCP SDK) for communication
  • 🆕 Style System v2 - Revolutionary type-safe style system (v0.4.0)
    • Every type has associated Style enum for compile-time registration
    • Runtime style selection with full type safety
    • Inline elicitation for all primitives with styled prompts
    • Auto-selection with silent defaults (zero ceremony)
    • Ergonomic builder pattern for one-off overrides
    • Extensible: define custom styles for any type (including built-ins)
  • 🆕 DateTime Support - Three major datetime libraries (v0.4.0)
    • chrono - DateTime, DateTime, NaiveDateTime
    • time - OffsetDateTime, PrimitiveDateTime
    • jiff - Timestamp, Zoned, civil::DateTime
    • Dual input methods: ISO 8601 strings or manual components
  • 🆕 JSON Elicitation - Dynamic JSON value construction (v0.4.0)
    • serde_json::Value with all JSON types
    • Recursive elicitation for arrays and objects
    • Depth limits to prevent infinite recursion

Supported Types

Primitives

  • Numeric: i8, i16, i32, i64, i128, isize, u8, u16, u32, u64, u128, usize, f32, f64
  • Text: String, bool
  • Time: std::time::Duration
  • Filesystem: std::path::PathBuf
  • Network: IpAddr, Ipv4Addr, Ipv6Addr, SocketAddr, SocketAddrV4, SocketAddrV6
  • DateTime (optional features):
    • chrono feature: DateTime<Utc>, DateTime<FixedOffset>, NaiveDateTime
    • time feature: OffsetDateTime, PrimitiveDateTime
    • jiff feature: Timestamp, Zoned, civil::DateTime
  • UUID (optional feature):
    • uuid feature: Uuid with parsing and random generation
  • JSON (optional feature):
    • serde_json feature: serde_json::Value (all JSON types)

Containers

  • Option: Option<T> - Optional values
  • Result: Result<T, E> - Success/failure outcomes
  • Vec: Vec<T> - Dynamic arrays
  • Arrays: [T; N] - Fixed-size arrays (any size N)
  • Tuples: (T1, T2, ...) - Heterogeneous tuples (up to arity 12)

Smart Pointers

  • Box: Box<T> - Heap allocation
  • Rc: Rc<T> - Reference counting
  • Arc: Arc<T> - Atomic reference counting

Collections

  • HashMap: HashMap<K, V> - Hash-based key-value map with duplicate key handling
  • BTreeMap: BTreeMap<K, V> - Ordered key-value map
  • HashSet: HashSet<T> - Hash-based unique set with automatic deduplication
  • BTreeSet: BTreeSet<T> - Ordered unique set
  • VecDeque: VecDeque<T> - Double-ended queue
  • LinkedList: LinkedList<T> - Doubly-linked list

Custom Types

  • Enums: Automatic Select paradigm via #[derive(Elicit)]
  • Structs: Automatic Survey paradigm via #[derive(Elicit)]
  • Nested: Unlimited nesting depth (e.g., Vec<Option<Result<HashMap<String, Arc<T>>, E>>>)

Installation

Add this to your Cargo.toml:

[dependencies]
elicitation = "0.3"
rmcp = "0.12"
tokio = { version = "1", features = ["macros", "rt-multi-thread"] }

Optional Features

Enable additional functionality with feature flags:

[dependencies]
elicitation = { version = "0.2", features = ["chrono", "time", "jiff", "serde_json"] }

Available features:

  • chrono - Enable chrono datetime types
  • time - Enable time datetime types
  • jiff - Enable jiff datetime types
  • serde_json - Enable JSON Value elicitation
  • api - Empty marker for API integration tests

MCP Setup

This library requires an MCP client (like Claude Desktop or Claude CLI) to provide the elicitation tools. Your application runs as an MCP server that the client invokes.

Running with Claude CLI

To run the examples or your own code:

# Install Claude CLI if you haven't already
# (see https://docs.anthropic.com/en/docs/agents-and-tools)

# Run an example through Claude CLI
claude-cli mcp add elicitation-demo --command "cargo run --example structs"

# Or ask Claude to run it
claude "Run the structs example from the elicitation crate"

Integration with Claude Desktop

Add your MCP server to Claude Desktop's configuration:

macOS: ~/Library/Application Support/Claude/claude_desktop_config.json Windows: %APPDATA%\Claude\claude_desktop_config.json Linux: ~/.config/claude/claude_desktop_config.json

{
  "mcpServers": {
    "elicitation-app": {
      "command": "/path/to/your/binary",
      "args": [],
      "env": {}
    }
  }
}

How It Works

  1. Your application creates an MCP client with rmcp::transport::stdio()
  2. Claude (the MCP client) provides elicitation tools via stdin/stdout
  3. When you call .elicit(), it sends tool requests to Claude
  4. Claude prompts the user and validates responses
  5. Your code receives strongly-typed Rust values

Note: Examples won't work standalone - they must be invoked by an MCP client.

Quick Start

use elicitation::{Elicit, Elicitation, ElicitResult};
use rmcp::ServiceExt;

// Derive for enums (Select pattern)
#[derive(Debug, Elicit)]
#[prompt("Choose your priority level:")]
enum Priority {
    Low,
    Medium,
    High,
    Critical,
}

// Derive for structs (Survey pattern)
#[derive(Debug, Elicit)]
struct Task {
    #[prompt("What's the task title?")]
    title: String,

    #[prompt("Describe the task:")]
    description: String,

    priority: Priority,

    #[prompt("Estimated hours (optional):")]
    estimated_hours: Option<i32>,
}

#[tokio::main]
async fn main() -> ElicitResult<()> {
    // Create MCP client via stdio transport
    let client = ()
        .serve(rmcp::transport::stdio())
        .await?;

    // Elicit a complete task from the user
    let task = Task::elicit(&client).await?;

    println!("Created task: {:?}", task);

    Ok(())
}

Examples

All examples require an MCP client (Claude Desktop or Claude CLI) to run. See MCP Setup above.

Primitive Types

// Elicit basic Rust types
let age: i32 = i32::elicit(&client).await?;
let name: String = String::elicit(&client).await?;
let confirmed: bool = bool::elicit(&client).await?;
let nickname: Option<String> = Option::<String>::elicit(&client).await?;
let scores: Vec<i32> = Vec::<i32>::elicit(&client).await?;

// Result types for success/failure outcomes
let operation: Result<String, i32> = Result::elicit(&client).await?;

Try it: claude "Run the simple_types example" or claude "Run the result example"

Filesystem Paths

use std::path::PathBuf;

// Elicit a filesystem path
let file_path: PathBuf = PathBuf::elicit(&client).await?;

// Optional paths work too
let config_path: Option<PathBuf> = Option::<PathBuf>::elicit(&client).await?;

Try it: claude "Run the pathbuf example"

Network Addresses

use std::net::{IpAddr, Ipv4Addr, Ipv6Addr, SocketAddr};

// Elicit IP addresses with automatic validation
let ip: IpAddr = IpAddr::elicit(&client).await?; // IPv4 or IPv6
let ipv4: Ipv4Addr = Ipv4Addr::elicit(&client).await?; // IPv4 only
let ipv6: Ipv6Addr = Ipv6Addr::elicit(&client).await?; // IPv6 only

// Socket addresses (IP + port)
let socket: SocketAddr = SocketAddr::elicit(&client).await?;

Try it: claude "Run the network example"

Time Durations

use std::time::Duration;

// Elicit duration in seconds (supports decimals)
let timeout: Duration = Duration::elicit(&client).await?;

// Works with optional durations
let cache_ttl: Option<Duration> = Option::<Duration>::elicit(&client).await?;

// Collections of durations
let intervals: Vec<Duration> = Vec::<Duration>::elicit(&client).await?;

Try it: claude "Run the duration example"

Enums (Select Pattern)

Unit Variants (Simple Selection)

#[derive(Debug, Elicit)]
enum Status {
    Pending,
    InProgress,
    Completed,
}

let status = Status::elicit(&client).await?;

Tuple Variants (Select + Field Elicitation)

#[derive(Debug, Elicit)]
enum MediaSource {
    Url(String),
    Base64(String),
    Binary(Vec<u8>),
}

// User first selects variant (Url/Base64/Binary), then provides the field value
let source = MediaSource::elicit(&client).await?;

Struct Variants (Select + Multi-Field Survey)

#[derive(Debug, Elicit)]
enum Input {
    Text(String),
    Image {
        mime: Option<String>,
        source: MediaSource,
    },
    Document {
        format: String,
        content: String,
    },
}

// User selects variant, then provides each field
let input = Input::elicit(&client).await?;

Mixed Variants

All three variant types can coexist in the same enum:

#[derive(Debug, Elicit)]
enum Status {
    Ok,                                     // Unit variant
    Warning(String),                        // Tuple variant
    Error { code: i32, msg: String },      // Struct variant
}

let status = Status::elicit(&client).await?;

Try it: claude "Run the enums example"

Structs (Survey Pattern)

#[derive(Debug, Elicit)]
struct Person {
    #[prompt("What's your name?")]
    name: String,

    #[prompt("How old are you?")]
    age: u8,

    #[prompt("What's your email?")]
    email: String,
}

let person = Person::elicit(&client).await?;

Try it: claude "Run the structs example"

Complex Nested Types

#[derive(Debug, Elicit)]
struct Project {
    name: String,
    team: Vec<Member>,
    tasks: Vec<Task>,
    budget: Option<f64>,
}

let project = Project::elicit(&client).await?;

Try it: claude "Run the complex_survey example"

Collections

use std::collections::{HashMap, HashSet};

// Elicit a HashMap with duplicate key handling
let scores: HashMap<String, i32> = HashMap::elicit(&client).await?;

// Elicit a HashSet with automatic deduplication
let tags: HashSet<String> = HashSet::elicit(&client).await?;

// BTreeMap and BTreeSet also supported for ordered collections
use std::collections::{BTreeMap, BTreeSet};
let config: BTreeMap<String, String> = BTreeMap::elicit(&client).await?;
let priorities: BTreeSet<i32> = BTreeSet::elicit(&client).await?;

// VecDeque and LinkedList for sequential access patterns
use std::collections::{VecDeque, LinkedList};
let queue: VecDeque<String> = VecDeque::elicit(&client).await?;
let linked: LinkedList<i32> = LinkedList::elicit(&client).await?;

Try it: claude "Run the collections example"

DateTime Types (Feature Flags)

Enable datetime support with feature flags: chrono, time, or jiff.

Chrono

use chrono::{DateTime, Utc, FixedOffset, NaiveDateTime};

// Elicit timezone-aware datetime (UTC)
let timestamp: DateTime<Utc> = DateTime::<Utc>::elicit(&client).await?;

// Elicit timezone-aware datetime (with offset)
let event: DateTime<FixedOffset> = DateTime::<FixedOffset>::elicit(&client).await?;

// Elicit naive datetime (no timezone)
let meeting: NaiveDateTime = NaiveDateTime::elicit(&client).await?;

Time

use time::{OffsetDateTime, PrimitiveDateTime};

// Elicit timezone-aware datetime
let event: OffsetDateTime = OffsetDateTime::elicit(&client).await?;

// Elicit datetime without timezone
let schedule: PrimitiveDateTime = PrimitiveDateTime::elicit(&client).await?;

Jiff

use jiff::{Timestamp, Zoned, civil::DateTime};

// Elicit Unix timestamp
let ts: Timestamp = Timestamp::elicit(&client).await?;

// Elicit timezone-aware datetime (DST-aware!)
let event: Zoned = Zoned::elicit(&client).await?;

// Elicit calendar datetime (no timezone)
let meeting: DateTime = DateTime::elicit(&client).await?;

Dual input methods: All datetime types support both ISO 8601 strings OR manual component entry (year, month, day, etc.).

JSON Values (Feature Flag)

Enable with serde_json feature flag to elicit dynamic JSON structures:

use serde_json::Value;

// Elicit any JSON value (null, bool, number, string, array, object)
let config: Value = Value::elicit(&client).await?;

// Works with nesting
let nested: Vec<Value> = Vec::<Value>::elicit(&client).await?;
let optional: Option<Value> = Option::<Value>::elicit(&client).await?;

The elicitation process handles all JSON types recursively:

  • null - Explicit null value
  • bool - Boolean true/false
  • number - Any JSON number
  • string - Text value
  • array - List of JSON values (recursive)
  • object - Key-value map (recursive)

Depth limit of 10 prevents infinite recursion.

Style System 🎨 (v0.2.2)

Revolutionary feature: Customize prompts per field with multiple styles!

#[derive(Debug, Elicit)]
struct Config {
    // Multiple prompt styles for same field
    #[prompt("Name", style = "curt")]
    #[prompt("What is your full name?", style = "verbose")]
    name: String,
    
    #[prompt("Age?", style = "curt")]
    #[prompt("Please enter your age in years", style = "verbose")]
    age: u32,
    
    // Mix styled and default prompts
    #[prompt("Enter city")]  // Used when style doesn't have override
    #[prompt("City", style = "curt")]
    city: String,
}

How it works:

  1. Collect unique style names from all #[prompt(..., style = "name")] attributes
  2. Generate ConfigElicitStyle enum with Default + collected styles
  3. At runtime, LLM or user selects style (just another Select elicitation!)
  4. Each field uses its style-specific prompt (or falls back to default)

Style selection is a state machine step - irrelevant whether LLM or user chooses. The style system separates what to ask (behavior) from how to ask (presentation).

Built-in styles available (for programmatic use):

  • DefaultStyle - Standard prompts
  • CompactStyle - Terse, minimal prompts
  • VerboseStyle - Detailed, explanatory prompts
  • WizardStyle - Step-by-step with progress

Currently, only String fields support styled prompts with inline elicitation. Other types fall back to default elicitation (support expanding in future versions).

Interaction Paradigms

Select

For choosing from a finite set of options (enums):

#[derive(Elicit)]
#[prompt("Choose your programming language:")]
enum Language {
    Rust,
    Python,
    JavaScript,
}

Affirm

For yes/no questions (booleans):

let confirmed: bool = bool::elicit(&client).await?;

Survey

For multi-field data collection (structs):

#[derive(Elicit)]
#[prompt("Let's create your profile:")]
struct Profile {
    name: String,
    age: u8,
    bio: Option<String>,
}

Authorize

Permission-based elicitation (planned for v0.2.0).

Attributes

#[prompt("...")]

Customize prompts for types or fields:

#[derive(Elicit)]
#[prompt("Configure your account:")] // Struct-level prompt
struct Account {
    #[prompt("Choose a username:")] // Field-level prompt
    username: String,
}

#[skip]

Skip fields during elicitation (uses Default::default()):

#[derive(Default, Elicit)]
struct Task {
    title: String,

    #[skip] // Not elicited, uses Default
    created_at: DateTime<Utc>,
}

Error Handling

The library provides rich error handling with location tracking:

use elicitation::{ElicitError, ElicitErrorKind, ElicitResult};

match Task::elicit(&client).await {
    Ok(task) => println!("Created: {:?}", task),
    Err(e) => eprintln!("Error: {}", e),
}

Error types:

  • InvalidFormat - Parsing failed
  • OutOfRange - Value outside valid range
  • InvalidOption - Invalid enum selection
  • MissingField - Required field missing
  • Cancelled - User cancelled operation
  • Mcp - MCP protocol error
  • Json - JSON parsing error

Architecture

Traits

  • Prompt - Provides prompt text for a type
  • Elicit - Implements elicitation logic
  • Select - For enum types (finite choices)
  • Affirm - For boolean types (yes/no)
  • Survey - For struct types (multi-field)

Type Composition

All elicitation types compose freely:

// Nested structures
let data: Vec<Option<Task>> = Vec::elicit(&client).await?;

// Complex hierarchies
#[derive(Elicit)]
struct Organization {
    name: String,
    departments: Vec<Department>,
}

#[derive(Elicit)]
struct Department {
    name: String,
    members: Vec<Member>,
    projects: Vec<Project>,
}

MCP Integration

The library uses the official rmcp (Rust MCP SDK) for MCP communication:

use rmcp::ServiceExt;

// Create client via stdio transport (for Claude Desktop/CLI)
let client = ()
    .serve(rmcp::transport::stdio())
    .await?;

// Use with elicitation
let value = MyType::elicit(&client).await?;

Documentation

Contributing

Contributions are welcome! Please:

  1. Fork the repository
  2. Create a feature branch (git checkout -b feature/amazing-feature)
  3. Commit your changes (git commit -m 'Add amazing feature')
  4. Push to the branch (git push origin feature/amazing-feature)
  5. Open a Pull Request

Development

# Run all checks
just check-all

# Run tests
cargo test --all

# Check examples
cargo check --examples

# Build documentation
cargo doc --open

Versioning

This project follows Semantic Versioning.

Current version: 0.2.0

License

Licensed under either of:

at your option.

Contribution

Unless you explicitly state otherwise, any contribution intentionally submitted for inclusion in the work by you, as defined in the Apache-2.0 license, shall be dual licensed as above, without any additional terms or conditions.

Acknowledgments

  • Built with rmcp - the official Rust MCP SDK
  • Powered by tokio for async runtime
  • Uses tracing for observability

About

Conversational elicitation of strongly-typed Rust values via MCP

Resources

License

Unknown, MIT licenses found

Licenses found

Unknown
LICENSE-APACHE
MIT
LICENSE-MIT

Stars

Watchers

Forks

Packages

No packages published

Contributors 2

  •  
  •