Skip to main content
Glama

Textile

CI License Python 3.15+ Code style: Ruff Type checked: Pyright

Layered Linux automation runtime, security supervisor, and Model Context Protocol (MCP) server for desktop intelligence.


Architecture

  • Twill (MCP Server): Exposes active capability strands over Model Context Protocol (stdio) to external assistants, clients, and IDEs.

  • Loom & Skein (Orchestration Engine): Compiles and dispatches capability strands across a prioritized layer hierarchy (0 to 1000). Higher numeric layers deterministically override lower layers.

  • Elastic (Event Fabric): Low-latency cross-process event bus featuring urgency-based filtering (AMBIENT, NOTICE, ALERT, FLASH), retained slot state storage, and SQLite WAL IPC synchronization.

  • Sensory Tapestry: System blackboard recording real-time ambient status, process metadata, and tool execution history.

  • Bubblewrap Sandboxing: Kernel namespace isolation for strand execution, enforcing read-only mounts, network restrictions, and workspace confinement.

  • TOTP 2FA Verification: RFC 6238 time-based one-time password gate protecting MUTATE, PRIVILEGED, and SYSTEM_EXEC execution tiers against unauthorized or accidental mutation.


Related MCP server: computer-use

Prerequisites

  • Python 3.15+ and uv

  • Bubblewrap (bwrap) (Required for sandboxed strand execution):

# Arch Linux
sudo pacman -S bubblewrap

# Debian / Ubuntu
sudo apt install bubblewrap

# Fedora
sudo dnf install bubblewrap

Quick Start

1. Run the MCP Server (stdio)

Run the Model Context Protocol stdio interface for external clients:

uv run textile twill

2. Manage 2FA Authenticator (RFC 6238 TOTP)

Generate setup secrets and current authentication codes:

# Display 2FA setup URI and QR information
uv run textile 2fa

# Print current 6-digit verification code
uv run textile 2fa -c

3. CLI Execution and Inspection

# List all registered strands
uv run textile strands

# Filter strands by capability tier
uv run textile strands --tier observe

# Inspect parameter schema and documentation for a strand
uv run textile inspect hyprland_focus_workspace

# Execute a strand directly
uv run textile call clipboard_set text="Textile ready"
uv run textile call clipboard_get

# Execute a privileged strand using 2FA
uv run textile call hyprland_exit_session --otp auto

4. Runtime and Yarn Inspection

# Inspect layer hierarchy, active yarns, and capability overrides
uv run textile skein

# List discovered yarn packages and search paths
uv run textile yarns

# View engine task ledger and blackboard status
uv run textile tapestry

5. User Settings Configuration

Per-yarn user configurations are managed in ~/.config/textile/settings.toml:

# Generate documented configuration template
uv run textile settings init

# View settings schema for a specific yarn
uv run textile settings docs hyprland

Capability Tiers and Security Policy

Every strand declares an explicit capability tier governing execution privileges:

Tier

Policy Gate

Sandboxing Profile

Typical Operations

OBSERVE

Open (No 2FA)

Read-only mounts

Telemetry queries, sensor reads, window state

INTERACT

Open (No 2FA)

Workspace / Network mounts

Desktop notifications, window focus, clipboard

MUTATE

2FA Required

Isolated write mounts

Persistent file modifications, config updates

PRIVILEGED

2FA Required

Host execution boundary

Session exit, power management, system reboots

SYSTEM_EXEC

2FA Required

Full host access

Package management, kernel modules, system-level execution


Discovery Search Paths

Yarns are discovered dynamically using ordered XDG-compliant search locations:

  1. TEXTILE_YARN_PATH environment variable override (colon-delimited)

  2. User configuration directory: ~/.config/textile/yarns/

  3. User installed data directory: ~/.local/share/textile/yarns/

  4. System-wide installed directories: /usr/local/share/textile/yarns/, /usr/share/textile/yarns/


Authoring Capability Yarns

To author new capability packages, use the standalone SDK: Needle.

uv add needle

See the Needle documentation for @strand and Yarn contracts.

Available Tools

46 tools
audit_yarn_integrityA

Audit system-wide yarn health, runtime dependencies, layer overrides, and schemas.

Inputs: no named parameters.

cancel_live_taskB

Cancel/terminate a currently running strand execution by its task ID or strand name.

Inputs: strand_name (required).

capture_screenB

Capture a high-resolution screenshot snapshot of the active screen and windows.

Inputs: purpose.

fetch_webpageB

Fetch and read the text content of a web page URL.

Inputs: url (required).

file_chmodB

Change permissions mode for a file or directory path.

Inputs: mode (required), path (required).

file_findA

Search for files matching a glob pattern in a directory path.

Inputs: path, pattern.

file_listB

List files and directories within a target directory path.

Inputs: path.

file_opC

File system operations, directory navigation, permissions, disk usage, and inotify monitoring.

Inputs: operation (required), mode, path, events, +8 more.

file_readA

Read contents of a text file with optional start and end line ranges.

Inputs: path (required), end_line, start_line.

file_replaceB

Replace exact text substring within a target file.

Inputs: path (required), target (required), replacement.

file_statA

Get stat metadata, permissions, size, and timestamps for a file path.

Inputs: path (required).

file_writeB

Write or overwrite text content to a target file path.

Inputs: path (required), content (required).

inotify_list_watchesA

List all active inotify watch descriptors and paths.

Inputs: no named parameters.

inotify_read_eventsB

Poll and read available inotify kernel events from active watches.

Inputs: max_events, timeout_ms.

inotify_unwatchA

Remove an active inotify watch by descriptor ID or path.

Inputs: watch (required).

inotify_wait_eventC

Wait synchronously for a specific filesystem event on a path.

Inputs: path (required), events, timeout_seconds.

inotify_watchC

Add a Linux inotify kernel watch on a file or directory.

Inputs: path (required), events, recursive.

launch_appC

Launch a desktop application or background command (routed through UWSM scope if active).

Inputs: app (required), is_tui.

packagekit_check_updatesB

Check for pending package and system software updates via PackageKit D-Bus.

Inputs: no named parameters.

packagekit_get_detailsC

Get detailed metadata, description, license, and repository info for a package via PackageKit D-Bus.

Inputs: package (required).

packagekit_installB

Install one or more packages non-interactively via PackageKit D-Bus with Polkit authorization.

Inputs: packages (required).

packagekit_refresh_cacheB

Refresh package manager repository metadata cache via PackageKit D-Bus.

Inputs: force.

packagekit_removeA

Remove one or more installed packages via PackageKit D-Bus.

Inputs: packages (required), autoremove.

packagekit_what_providesA

Find which package provides a specific file path or binary via PackageKit D-Bus.

Inputs: file_path (required).

polkit_check_authB

Check authorization for a Polkit action ID against org.freedesktop.PolicyKit1.Authority.

Inputs: action_id (required), details, allow_user_interaction.

polkit_generate_policyC

Generate standards-compliant .policy XML definitions for custom actions.

Inputs: actions (required), vendor, vendor_url, output_path.

polkit_generate_ruleB

Generate a Polkit-1 JavaScript rule (.rules) for pre-authorizing specific actions.

Inputs: rule_name (required), action_pattern (required), users, groups, +2 more.

polkit_list_actionsA

List and search registered Polkit action definitions.

Inputs: filter_query.

polkit_pkexecB

Execute a command with elevated privileges using pkexec CLI escalation.

Inputs: command (required), user, timeout_seconds.

process_get_loadavgA

Get CPU 1, 5, and 15-minute load averages and CPU core counts.

Inputs: no named parameters.

process_get_priorityA

Get nice priority level of a running process by PID.

Inputs: pid (required).

process_killA

Send a POSIX signal to terminate or signal a process by PID.

Inputs: pid (required), signal.

process_listB

List running system processes with PID, CPU/memory usage, user, and command line.

Inputs: filter.

process_list_bg_jobsA

List background jobs spawned and tracked by Textile.

Inputs: no named parameters.

process_set_priorityB

Set nice priority level (-20 to 19) of a process by PID.

Inputs: pid (required), priority.

run_commandA

Execute a bash shell command for developer tasks (git, tests, builds, package managers).

Inputs: command (required), background.

run_system_testsA

Run the full Textile system diagnostic unit, integration, and E2E test suite.

Inputs: no named parameters.

search_webB

Search the live internet for documentation, code examples, API references, library versions, or error solutions.

Inputs: query (required).

sensors_get_cpu_freqsA

Retrieve live CPU core frequencies (MHz) and scaling governors across CPU cores.

Inputs: no named parameters.

sensors_get_telemetryA

Retrieve hardware telemetry (temperatures, fans, voltages, power) via lm_sensors / sysfs.

Inputs: no named parameters.

storage_disk_usageA

Get total, used, and available disk usage for a storage path.

Inputs: path.

storage_list_mountsB

List mounted filesystems and storage devices.

Inputs: no named parameters.

textile_get_engine_stateA

Get the core engine execution state (active running tasks and execution history).

Inputs: no named parameters.

textile_get_sensory_stateA

Get the open sensory blackboard snapshot (sensory state slots and recent stitched notices/alerts).

Inputs: no named parameters.

textile_get_stateA

Get full snapshot of the sensory blackboard state slots and notices.

Inputs: no named parameters.

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    TDQS

    C2.9/5.0

    Scored across 46 tools

    Disambiguation2/5

    Several tools have muddy boundaries: textile_get_state and textile_get_sensory_state describe nearly the same blackboard snapshot, and file_op is a catch-all overlapping file_list, file_find, file_stat, file_chmod, and inotify. run_command also overlaps with launch_app, polkit_pkexec, and process management, making selection error-prone.

    Naming Consistency4/5

    Most tools follow a clear domain_prefix + action pattern: packagekit_*, polkit_*, process_*, file_*, inotify_*, sensors_*, and textile_get_*. The main blemishes are file_op, a noun-style exception, and a handful of unprefixed tools like search_web, launch_app, and run_command, but overall the convention is recognizable.

    Tool Count2/5

    46 tools is far beyond a typical well-scoped MCP server. The set spans files, processes, packages, polkit, inotify, sensors, web fetching, screen capture, tests, and internal Textile state, which feels like several specialized servers crammed into one.

    Completeness3/5

    Coverage is broad but uneven: package management and polkit are quite complete, while file operations lack delete/move/copy tools and process management lacks a dedicated start tool. Some gaps can be papered over with run_command, but the surface has clear dead ends for a supposed system administration toolkit.

    Maintenance

    ActivityMaintained
    ResponsivenessNo recent community issues

    Related MCP Connectors

    Related MCP Servers

    • A
      license
      Not graded
      quality
      D
      maintenance
      Enables full Linux desktop control including windows, mouse, keyboard, clipboard, audio, screenshots, OCR, accessibility, and system management through MCP-compatible AI agents.
      1
      MIT
    • A
      license
      B
      quality
      D
      maintenance
      Exposes Anthropic's computer-use action surface (screenshot, click, move, keyboard, clipboard, batch) against a persistent desktop display via MCP stdio protocol. Enables AI agents to control a virtual desktop environment through natural language instructions.
      24
      MIT
    • A
      license
      Not graded
      quality
      B
      maintenance
      Enables controlling your own Linux desktop machines from any MCP-capable LLM client, with opt-in agent capabilities for shell, filesystem, processes, screenshots, system info, and input injection.
      1
      MIT
    • A
      license
      Not graded
      quality
      B
      maintenance
      Enables AI agents to automate a Linux desktop natively on Wayland, with 68 tools for screen capture, input simulation, browser automation, OCR, accessibility, media, audio, clipboard, notifications, filesystem, system, memory, and network operations through the MCP protocol.
      MIT