aether-mcp-servers 0.8.9

# mcp-servers

Pre-built [MCP](https://modelcontextprotocol.io/) tool servers for Aether agents. Each server runs in-process and is gated behind a feature flag so you only compile what you need.

| Feature | Server | What it provides |
|---------|--------|-----------------|
| `coding` | [`CodingMcp`](src/coding/README.md) | File read/write/edit, bash, LSP integration, web fetch/search |
| `skills` | [`SkillsMcp`](src/skills/README.md) | Slash commands and reusable skill prompts |
| `tasks` | [`TasksMcp`](src/tasks/README.md) | Hierarchical task management with dependencies |
| `subagents` | [`SubAgentsMcp`](src/subagents/README.md) | Spawn concurrent sub-agents in the foreground or as an MCP Task |
| `review` | [`ReviewMcp`](src/review/README.md) | Review agent-authored Markdown via native elicitation |

## Table of Contents

<!-- START doctoc generated TOC please keep comment here to allow auto update -->
<!-- DON'T EDIT THIS SECTION, INSTEAD RE-RUN doctoc TO UPDATE -->

- [Documentation](#documentation)
- [Using with Aether (mcp.json)](#using-with-aether-mcpjson)
- [Programmatic Usage](#programmatic-usage)
- [Server Documentation](#server-documentation)
- [Feature Flags](#feature-flags)

<!-- END doctoc generated TOC please keep comment here to allow auto update -->

## Documentation

Full API documentation is available on [docs.rs](https://docs.rs/aether-mcp-servers).

Key entry points:
- [`CodingMcp`] -- file I/O, shell, search, and LSP tools
- [`LspRegistry`](lsp::LspRegistry) -- manages LSP daemon connections
- [`TasksMcp`](tasks::TasksMcp) -- hierarchical task management
- [`SkillsMcp`](skills::SkillsMcp) -- skill prompts and slash commands
- [`SubAgentsMcp`](subagents::SubAgentsMcp) -- sub-agent orchestration
- [`ReviewMcp`](review::ReviewMcp) -- Markdown artifact review and feedback 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`:

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

Each server key must match a factory registered with `McpBuilder::register_in_memory_server()`. Loading the config only expands the server's variables. The factory is invoked exactly once during `McpBuilder::spawn`, after the deferred-tools endpoint is bound. It receives the server's `args`, with variables expanded, plus `RuntimeServices` containing the root directory, `AgentDeps`, and the deferred-tools socket path (when any server defers its tools). The `args` array is parsed as CLI flags by each server, with relative paths resolved against the root directory. Invalid `args` fail `spawn` with `McpSpawnError::ServerFactory` rather than falling back to defaults:

| Server | Flag | Default | Description |
|--------|------|---------|-------------|
| `coding` | `--root-dir <path>` | workspace root | Workspace root for LSP and file operations |
| `coding` | `--rules-dir <path>` (repeatable) | none | Explicit prompt directories for automatic read-triggered rules |
| `coding` | `--permission-mode <mode>` | `always-allow` | `always-allow`, `auto`, or `always-ask` |
| `coding` | `--disable-lsp` | enabled | Disable LSP-backed tools and `aether-lspd` daemon connections |
| `skills` | `--dir <path>` (repeatable) | required | Prompt directories to scan |
| `tasks` | `--dir <path>` | session-scoped temp dir | Base directory for persistent task storage (creates `.aether-tasks/` inside) |
| `subagents` | `--project-root <path>` (alias: `--dir`) | workspace root | Project root containing optional `.aether/settings.json` authored agents |
| `review` | `--root-dir <path>` | workspace root | Base for relative artifact paths |

To register factories and load the config:

```rust,ignore
use aether_core::mcp::mcp;
use mcp_servers::McpBuilderExt;
use aether_core::mcp::McpConfig;

let builder = mcp("/my/project")
    .with_agent_deps(deps)
    .with_builtin_servers()
    .with_config(McpConfig::from_json_file("mcp.json")?)?;

// Built-in factories have not run yet. They receive RuntimeServices here.
let runtime = builder.spawn()?;
```

Custom factories use the same lazy signature and return an error for invalid `args`:

```rust,ignore
builder.register_in_memory_server("custom", |args, services| {
    Ok(McpServer::new(CustomMcp::from_args(args, &services.root_dir)?))
});
```

## Programmatic Usage

Add to your `Cargo.toml`:

```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:

```rust,ignore
use mcp_servers::{CodingMcp, SkillsMcp, TasksMcp};
use mcp_utils::server::McpServer;

// Create a coding server
let server = McpServer::new(CodingMcp::new("/my/project"));

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

server.serve_stdio().await?;
```

## Server Documentation

- [`CodingMcp`](src/coding/README.md)
- [`SkillsMcp`](src/skills/README.md)
- [`TasksMcp`](src/tasks/README.md)
- [`SubAgentsMcp`](src/subagents/README.md)
- [`ReviewMcp`](src/review/README.md)

---

## Feature Flags

- **`default`** -- coding, skills, tasks, subagents, and review 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`, `review`)
- **`review`** -- Markdown artifact review through elicitation
- **`all`** -- explicit alias enabling all built-in servers