mcp-servers

Pre-built MCP tool servers for Aether agents. Each server runs in-process and is gated behind a feature flag so you only compile what you need.

Table of Contents

• Documentation
• Using with Aether (mcp.json)
• Programmatic Usage
• Server Documentation
• Feature Flags

Documentation

Full API documentation is available on docs.rs.

Key entry points:

• [CodingMcp] -- file I/O, shell, search, and LSP tools
• CodingTools -- trait for custom tool backends
• LspRegistry -- manages LSP daemon connections
• TasksMcp -- hierarchical task management
• SkillsMcp -- skill prompts and slash commands
• SubAgentsMcp -- sub-agent orchestration
• SurveyMcp -- structured user input collection
• PlanMcp -- plan review and approval workflow
• [McpBuilderExt] -- register all servers in one call

Using with Aether (mcp.json)

These servers use Aether's in-memory transport type -- they run inside your agent process, not as separate subprocesses. Wire them up in mcp.json:

{
  "servers": {
    "coding": {
      "type": "in-memory",
      "args": ["--rules-dir", ".aether/skills"]
    },
    "skills": {
      "type": "in-memory",
      "args": ["--dir", ".aether/skills", "--notes-dir", ".aether/notes"]
    },
    "tasks": {
      "type": "in-memory"
    },
    "subagents": {
      "type": "in-memory",
      "args": ["--project-root", "."]
    },
    "plan": {
      "type": "in-memory"
    }
  }
}

Each server key must match a factory registered with McpBuilder::register_in_memory_server(). The args array is parsed as CLI flags by each server:

To register factories and load the config:

use aether_core::mcp::mcp;
use futures::FutureExt;
use mcp_servers::{CodingMcp, PlanMcp, SkillsMcp, SubAgentsMcp, TasksMcp};

let builder = mcp()
    .register_in_memory_server("coding", Box::new(|_args| {
        async move { CodingMcp::new().into_dyn() }.boxed()
    }))
    .register_in_memory_server("skills", Box::new(|args| {
        async move { SkillsMcp::from_args(args).unwrap().into_dyn() }.boxed()
    }))
    .register_in_memory_server("tasks", Box::new(|args| {
        async move { TasksMcp::from_args(args).unwrap().into_dyn() }.boxed()
    }))
    .register_in_memory_server("subagents", Box::new(|args| {
        async move { SubAgentsMcp::from_args(args).unwrap().into_dyn() }.boxed()
    }))
    .register_in_memory_server("plan", Box::new(|_args| {
        async move { PlanMcp::new().into_dyn() }.boxed()
    }));

// Load mcp.json -- matches server keys to registered factories
let builder = builder.from_json_files(&["mcp.json"]).await?;

Programmatic Usage

Add to your Cargo.toml:

# Everything
mcp-servers = { path = "../mcp-servers" }

# Just coding tools
mcp-servers = { path = "../mcp-servers", default-features = false, features = ["coding"] }

Create and start servers directly:

use mcp_servers::{CodingMcp, SkillsMcp, TasksMcp};
use rmcp::ServiceExt;

// Create a coding server
let server = CodingMcp::new()
    .with_root_dir("/my/project".into())
    .into_dyn();

// Or with LSP support
let server = CodingMcp::new()
    .with_root_dir("/my/project".into())
    .with_lsp("/my/project".into())
    .into_dyn();

Server Documentation

• CodingMcp
• SkillsMcp
• TasksMcp
• SubAgentsMcp
• SurveyMcp
• PlanMcp

Feature Flags

• default -- coding, skills, tasks, subagents, and plan servers
• coding -- file ops, bash, LSP, web tools
• skills -- slash commands and prompts
• tasks -- task tracking (no dependency on coding)
• subagents -- sub-agent spawning (implies coding, skills, tasks, survey, plan)
• survey -- structured human elicitation tooling
• plan -- markdown plan submission and approval workflow
• all -- explicit alias enabling all built-in servers