Slide MCP Server

An MCP server implementation that integrates with the Slide API, designed around the questions an MSP technician actually asks Claude Desktop - "are all my Slide boxes healthy?", "did backups run last night for ACME?", "find Q4-budget.xlsx on Bob's laptop and restore Tuesday's version".

What to ask first

If you've just installed the extension, try one of these prompts. Claude will reach for slide-mcp-server automatically - you don't have to spell out "use the Slide MCP".

"Are all my Slide boxes healthy?"
"Did backups run last night for ACME?"
"Find Q4-budget.xlsx on Bob's laptop."
"What unresolved alerts do I have? Sort worst first."
"Boot a recovery VM for DC-01."
"What changed in the last 24 hours?"
"I don't know where to start - what can you do?"

You never need to know a_xxxxxxxxxxxx / d_xxxxxxxxxxxx IDs. Say "Bob's laptop" or "the file server" - the server fuzzy-matches against hostnames and display names and asks you to disambiguate when more than one entity matches.

When in doubt: /slide.welcome (slash-command in Claude Desktop) or just ask "what can you do?".

What's new in v5.0.0

Novice-first overhaul. The LLM now reaches for slide-mcp-server whenever the user mentions Slide, BCDR, backups, restores, snapshots, recovery VMs, file recovery, disaster recovery, audit logs, or unresolved alerts - without the user having to spell it out.
slide_help - new always-available tool with getting_started, examples, glossary, troubleshoot, what_can_you_do operations.
/slide.welcome slash-command prompt + slide://welcome, slide://help/glossary, slide://help/troubleshoot resources.
Server-side fuzzy name resolution. Every tool that takes an agent_id / device_id / client_id also accepts name_hint. Pass a hostname, display name, or substring; the server resolves it. Ambiguous matches return a structured list of candidates.
Startup token validation with clear, multi-line remediation guidance for invalid tokens or firewall issues.
slide-mcp-server --doctor subcommand for self-diagnosis (token, network, sample reads). Idempotent and CI-friendly.
next_steps array appended to every response with curated follow-up tool-call suggestions. Suppress with hints=off.
Friendlier errors: 401/403/404 carry novice-friendly remediation hints, and unknown operations get did-you-mean suggestions.

Full migration table and per-file detail in CHANGELOG.md.

What's new in v4.0.0

Headline new capability: slide_files. Search filenames across every snapshot for an agent, list snapshot versions of a path, then create a restore, browse it, or push files back to the protected system - all through one task-oriented tool. Wraps the previously-unexposed Slide API v1.27.0 FileSearch + PathVersion endpoints.
slide_audit for compliance / "what changed in the last 24h?" - wraps the previously-unexposed /v1/audit* endpoint family.
Task-oriented tool surface. New slide_overview (inventory + health + per-client/per-device summaries), slide_recovery (boot VMs + export images + DR networks), slide_clients, slide_admin.
5 MCP Prompts in Claude Desktop's slash-command UI: /slide.daily-status, /slide.triage-alerts, /slide.restore-file, /slide.boot-recovery-vm, /slide.dr-runbook.
9 MCP Resources (up from 1) including URI-templated slide://client/{id}, slide://device/{id}, slide://agent/{id}, plus static slide://overview/health, slide://alerts/unresolved, slide://audit/recent, and a live-cached slide://docs/openapi.
Token-conscious responses. Every list op defaults to format=summary (one-line entries) instead of v3's full JSON dump; opt up to compact or detailed when you need more, plus fields=a,b,c projection.
Tool annotations (readOnlyHint/destructiveHint/idempotentHint) so Claude Desktop skips confirmation prompts on safe reads.
Reliability: HTTP client now honours Retry-After on 429s, retries transient 5xx once, and surfaces structured APIErrors with per-status hints.
Permission tiers collapsed from four to three (read-only/safe/ full). Legacy names (reporting/restores/full-safe) are silently aliased.
Retired ~3K LoC of dead-weight tooling (slide_docs, slide_presentation, slide_reports, slide_meta) that the LLM can compose from raw data + the new prompts.

Full migration table and per-file detail in CHANGELOG.md.

🚀 Go Binary Implementation ⚡

Single binary: No dependencies, just download and run
Fast startup: ~50ms startup time
Low memory usage: 10-20MB memory footprint
Cross-platform: Linux, macOS (Apple Silicon + Intel), Windows
Task-oriented surface: 11 meta-tools designed around real MSP workflows
MCP Resources + Prompts: pre-baked context and slash-command workflows (slide://overview, /slide.daily-status, etc.)

⚡ Install

Pick the path for your host. Each is one step.

🟢 Claude Desktop - drag and drop (recommended)

Download slide-mcp-server.mcpb.
Drag it onto Claude Desktop's Settings -> Extensions screen.
Paste your Slide API token when prompted (generate one at console.slide.tech under My Settings -> API Tokens).

That's it. Claude Desktop owns the token storage, picks the right binary for your OS, and starts the server. The bundle is signed and notarized, so Gatekeeper won't get in the way on macOS.

🟦 Claude Code

claude mcp add slide --env SLIDE_API_KEY=tk_... -- /path/to/slide-mcp-server

Or, from a clone of this repo:

make build
SLIDE_API_KEY=tk_... make install-claude-code

The wrapper is idempotent - re-running it overwrites the prior slide registration cleanly.

🟪 Cursor / other MCP hosts

Drop the binary somewhere on $PATH and add this to ~/.cursor/mcp.json (or your host's equivalent):

{
  "mcpServers": {
    "slide": {
      "command": "slide-mcp-server",
      "env": { "SLIDE_API_KEY": "${SLIDE_API_KEY}" }
    }
  }
}

Claude Code, Cursor, and most other modern hosts support ${VAR} expansion so the literal token never has to live in the config file.

🛠️ From source / on a new dev box

git clone https://github.com/austinmcchord/slideMCP.git && cd slideMCP
make setup-dev   # idempotently installs go, jq, gh via Homebrew
make build       # produces build/slide-mcp-server
make pack-dxt    # produces build/slide-mcp-server.mcpb (unsigned, dev)

🔏 Code signing setup (release builds only)

make pack-dxt-signed (and the release scripts) need a valid Apple Developer ID Application certificate plus a notarytool keychain profile. A first-time setup helper handles the whole dance - generates a CSR, opens the Apple Developer portal in your browser, imports the resulting certificate, stores notarytool credentials, and writes scripts/release-env.sh (gitignored) so subsequent builds Just Work:

./scripts/setup-signing.sh    # interactive, idempotent, ~5 minutes
make doctor-signing            # read-only check; run before each release

Once setup is done, on every new shell:

source scripts/release-env.sh   # exports DEVELOPER_ID, KEYCHAIN_PROFILE, etc.
make pack-dxt-signed            # signed + notarized .mcpb in build/
make verify-dxt                 # confirms Gatekeeper accepts the inner binary

The setup script is not tracked in git (covered by the existing **/setup-signing.sh rule) and never writes plaintext secrets outside of the local scripts/release-env.sh (also gitignored) and your macOS keychain.

Read docs/SIGNING.md before touching anything in the signing pipeline. It documents 10+ macOS-specific quirks we already debugged the hard way (DER vs PEM, PKCS#1 vs PKCS#8, why .p12 imports fail on modern macOS, why add-trusted-cert -r trustAsRoot breaks chain validation, why spctl --type execute always rejects raw Mach-O even when notarized, and more). LLMs working on this repo: also see AGENTS.md.

For more setup detail (CLI flags, environment variables, manual JSON snippets), see the installation section further down.

🎯 Architecture

Task-oriented meta-tools: 11 focused meta-tools designed around the questions an MSP technician actually asks Claude Desktop, not the raw API endpoint surface. Each accepts an operation parameter that selects the action; conditional allOf/if/then schemas validate per-operation required parameters before the call ever leaves Claude.

Plus MCP Prompts (slash-command workflows in Claude Desktop) and URI-templated Resources (cheap pre-baked context the LLM can pull without a tool call).

Example Usage Pattern

{
  "name": "slide_files",
  "arguments": {
    "operation": "search",
    "agent_id": "a_0123456789ab",
    "search_term": "Q4-budget"
  }
}

Available Meta-Tools (v5.0.0)

💡 Help & onboarding

slide_help — discovery / onboarding / troubleshooting / debug
- Operations: getting_started, examples, glossary, troubleshoot, list_prompts, list_resources, what_can_you_do, debug
- Always available, never blocked by tools mode, never disable-able.
- First thing the LLM should call when the user is vague.
- debug returns a full diagnostic dump (config, DNS, TLS, live API probes, recent logs) — call it whenever a tool fails.

🧭 Overview & inventory

slide_overview — top-level inventory + health
- Operations: inventory, health, for_client, for_device
- One-screen answer to "what do we have?" and "is everything OK?"
- Read-only; safe to call at the start of any conversation.

🔎 Files (the headline v4 capability)

slide_files — search filenames + restore + push back
- Operations: search, versions, list_restores, get_restore, create_restore, delete_restore, browse, list_pushes, create_push, update_push, get_push_status
- search finds a filename across every snapshot for an agent (Slide API v1.27.0 FileSearch); versions lists snapshots that contain a path. The whole "find Q4-budget.xlsx and restore Tuesday's copy" flow lives in this one tool.

🛟 Recovery

slide_recovery — boot VMs, export images, manage DR networks
- VM operations: list_vms, get_vm, boot_vm, update_vm, delete_vm, get_rdp_bookmark
- Image exports: list_images, get_image, export_image (VHD/VHDX/VMDK/QCOW2/RAW), delete_image, browse_image
- DR networks: list_networks, get_network, create_network, update_network, delete_network plus IPSec / port-forward / WireGuard peer CRUD
- The "I need to actually recover something" toolkit.

📜 Audit (compliance / "what changed?")

slide_audit — account audit log
- Operations: list, get, actions, resources, recent
- recent defaults to a 24h window for "what just changed?"

🗂️ Snapshots, backups, alerts, agents, devices

slide_snapshots — list, list_deleted, get, get_service_verification, recent_for_agent (one-call "last N days of restore points for X").
slide_backups — list, get, start, plus the v4 status_for_client / status_for_device / recent_for_agent ops that answer "did backups run last night for X?" in one call.
slide_alerts — list, get, update, plus v4 triage that sorts unresolved alerts by severity hint.
slide_agents — full v1.27.0 surface (services, schedule, retention, restore-defaults, volumes, alert configs).
slide_devices — list, get, update, poweroff, reboot plus device network + per-device VLAN CRUD.

👥 Clients & admin

slide_clients — client CRUD (list, get, create, update, delete).
slide_admin — users + accounts + user avatar (list_users, get_user, get_user_avatar, list_accounts, get_account, update_account).

🔁 Backward-compat shim

list_all_clients_devices_and_agents still works; it forwards to slide_overview operation=inventory.

MCP Prompts (slash-command workflows)

Claude Desktop surfaces these in its slash-command UI:

/slide.welcome — one-message intro for first-time users
/slide.daily-status [client?] [hours?] — daily ops summary
/slide.triage-alerts — prioritised unresolved-alert review
/slide.restore-file [filename?] [agent?] — guided file recovery
/slide.boot-recovery-vm [agent?] — guided VM recovery
/slide.dr-runbook [device?] — DR runbook generated from the device's real agents and snapshots

MCP Resources (cheap pre-baked context)

Claude can read these without burning a tool call:

slide://welcome — one-page primer + example questions (no API calls)
slide://help/glossary — Slide-specific terminology (no API calls)
slide://help/troubleshoot — common setup / auth / network problems
slide://overview/inventory — clients → devices → agents tree
slide://overview/health — one-line-per-device-and-agent summary
slide://alerts/unresolved — open alerts, prioritised
slide://audit/recent — last 24h of audit log
slide://docs/openapi — live Slide OpenAPI spec (cached 1h)
slide://client/{client_id} — single client detail
slide://device/{device_id} — single device detail
slide://agent/{agent_id} — single agent detail
slide://agent/{agent_id}/snapshots/recent — recent snapshots

The legacy slide://context/clients-devices-agents URI keeps working as an alias for slide://overview/inventory.

Token-conscious response shape

Every list/get operation accepts these cross-cutting parameters:

format — summary (default for lists; one-line entries), compact (full payload, no indentation), or detailed (full payload, indented — equivalent to v3 behaviour).
fields — comma-separated JSON projection, e.g. fields=id,hostname,last_seen drops everything else.
hints — set to off to suppress the next_steps array v5 appends to most responses. Default is on; the hints are short and cheap.
All list responses include pagination.next_offset and count.

name_hint: no IDs to memorise

Every tool that needs an agent_id, device_id, or client_id also accepts name_hint. Pass a hostname, display name, client name, or any substring (name_hint=bob, name_hint=ACME, name_hint=DC-01) and the server fuzzy-resolves it server-side.

0 matches → the response is a {"name_hint_error":"no_match",...} payload with a suggestion to call slide_overview operation=inventory.
1 match → the resolved entity appears in the response under _resolved so the assistant can confirm to the user which one it picked, then the operation proceeds normally.
2+ matches → the response is a {"name_hint_error":"ambiguous","candidates":[...]} payload with up to 10 candidates. The assistant should ask the user to pick one, then re-call with the explicit *_id.

Key Features

🔐 Infrastructure Management

Device Control: Remote power operations, hostname management, client assignment
Agent Deployment: Automated pairing, display name management
Network Isolation: Client-specific networks with VPN access
Advanced Networking: IPSec tunnels, port forwarding, WireGuard peers

💽 Data Protection & Recovery

Automated Backups: Agent-based backup initiation and monitoring
Point-in-Time Recovery: Snapshot browsing with location filtering
Flexible Restores: File-level and full disk image exports
Multiple Formats: VHD, VHDX (dynamic/fixed), Raw disk images
Boot Modifications: Optional passwordless admin account creation

☁️ Virtualization

VM Creation: Create VMs from any snapshot
Resource Control: Configurable CPU/RAM allocation
Network Integration: Connect VMs to isolated networks
Console Access: Browser-based VNC for direct VM interaction
RDP Bookmarks: Generate downloadable .rdp files for easy Windows Remote Desktop access

📊 Monitoring & Organization

Alert Management: Centralized alert monitoring and resolution
Client Organization: Group resources by client for better management
User Management: Account access and permissions
Comprehensive Filtering: Advanced pagination and sorting across all resources

📋 Professional Data Presentation

Smart Cards: Individual and table-based cards for agents, clients, devices, and snapshots
Report Templates: Runbook procedures, daily summaries, and monthly analysis reports
Multiple Formats: HTML, Markdown, and HAML output for different use cases
Dashboard Ready: Pre-formatted cards perfect for status displays and monitoring
Documentation Support: Professional templates for operational procedures and troubleshooting

All meta-tools support pagination (limit, offset), sort options where applicable, and the cross-cutting format / fields parameters described above.

📦 Detailed installation reference

The "⚡ Install" section near the top of this README covers the recommended paths. This section documents the underlying configuration so you can hand-roll setups for unusual hosts or CI.

Getting an API Key

Log in to your Slide account
Navigate to My Settings -> API Tokens
Generate your API token. Treat it like a password - it grants full access to your Slide account.

Pre-built downloads

The headline release asset is the .mcpb (Claude Desktop Extension):

slide-mcp-server.mcpb (one file, every platform): https://github.com/austinmcchord/slideMCP/releases/latest/download/slide-mcp-server.mcpb

Standalone binaries (for claude mcp add / Cursor / CI):

macOS (universal Apple Silicon + Intel): slide-mcp-server-vX.Y.Z-darwin-universal.tar.gz
Linux x64: slide-mcp-server-vX.Y.Z-linux-amd64.tar.gz
Linux ARM64: slide-mcp-server-vX.Y.Z-linux-arm64.tar.gz
Windows x64: slide-mcp-server-vX.Y.Z-windows-x64.zip

All macOS artifacts are signed with a Developer ID and notarized.

Build from source

git clone https://github.com/austinmcchord/slideMCP.git
cd slideMCP
make build      # build/slide-mcp-server (current platform)
make pack-dxt   # build/slide-mcp-server.mcpb (universal bundle, unsigned)

Manual Claude Desktop config

If you can't use the .mcpb (older Claude Desktop, locked-down corp machine, etc.), edit claude_desktop_config.json directly:

{
  "mcpServers": {
    "slide": {
      "command": "/path/to/slide-mcp-server",
      "env": { "SLIDE_API_KEY": "YOUR_API_KEY_HERE" }
    }
  }
}

With non-default tool mode and disabled tools:

{
  "mcpServers": {
    "slide": {
      "command": "/path/to/slide-mcp-server",
      "env": {
        "SLIDE_API_KEY": "YOUR_API_KEY_HERE",
        "SLIDE_TOOLS": "read-only",
        "SLIDE_DISABLED_TOOLS": "slide_recovery"
      }
    }
  }
}

Or pass everything as CLI args:

{
  "mcpServers": {
    "slide": {
      "command": "/path/to/slide-mcp-server",
      "args": [
        "--api-key", "YOUR_API_KEY_HERE",
        "--tools", "full-safe",
        "--disabled-tools", "slide_agents,slide_backups"
      ]
    }
  }
}

Test Your Installation

Use the official MCP Inspector to drive the server interactively:

export SLIDE_API_KEY=tk_...
npx @modelcontextprotocol/inspector ./slide-mcp-server

Or run the bundled smoke test (covers initialize, tools/list, resources/list, the v1.27.0 validation paths, and a sampling of live tool calls):

SLIDE_API_KEY=tk_... ./smoke_test.sh

🔧 CLI Arguments & Configuration

The Slide MCP Server supports several command-line arguments for flexible configuration:

Command Line Arguments

# Basic usage with API key
./slide-mcp-server --api-key YOUR_API_KEY

# All available flags
./slide-mcp-server [OPTIONS]

Flag	Description	Environment Variable	Default
`--api-key`	Slide API key for authentication	`SLIDE_API_KEY`	Required
`--base-url`	Base URL for Slide API endpoint	`SLIDE_BASE_URL`	`https://api.slide.tech`
`--tools`	Permission mode (`read-only` / `safe` / `full`; legacy aliases `reporting`/`restores`/`full-safe` still work)	`SLIDE_TOOLS`	`safe`
`--disabled-tools`	Comma-separated list of tools to disable (`slide_help` cannot be disabled)	`SLIDE_DISABLED_TOOLS`	None
`--doctor`	Run self-diagnostic checks (token, network, sample reads) and exit	-	-
`--debug`	Dump a full diagnostic bundle (version, runtime, config, env, DNS, TLS, live API probes, recent logs) as JSON and exit. Safe to paste into a support thread; API token is masked. Works without an API key configured.	-	-
`--skip-startup-validation`	Skip the startup probe of `/v1/account` (useful when launching offline)	-	-
`--tool`	Run a single tool then exit (one-shot mode)	-	-
`--args`	JSON arguments for `--tool`	-	-
`--version`	Show version information and exit	-	-

Priority: CLI flags take precedence over environment variables.

Examples

# Using CLI flags (default 'safe' mode)
./slide-mcp-server --api-key sk_test_123 --base-url https://custom.api.endpoint

# Read-only mode (only list/get/search/browse)
./slide-mcp-server --api-key sk_test_123 --tools read-only

# Using environment variables
export SLIDE_API_KEY="sk_test_123"
export SLIDE_TOOLS="read-only"
./slide-mcp-server

# Mixed usage (CLI overrides environment)
export SLIDE_TOOLS="full"
./slide-mcp-server --api-key sk_test_123 --tools read-only

# Disable specific tools
./slide-mcp-server --api-key sk_test_123 --disabled-tools "slide_recovery,slide_files"

# One-shot tool invocation (handy for scripts / smoke tests)
./slide-mcp-server --api-key sk_test_123 --tool slide_overview --args '{"operation":"health"}'

# Show version
./slide-mcp-server --version
# Output: slide-mcp-server version 4.0.0

🚫 Disabling Specific Tools

In addition to permission modes, you can disable specific tools entirely using the --disabled-tools flag or SLIDE_DISABLED_TOOLS environment variable. This provides fine-grained control over which tools are available.

Usage Examples

# Disable specific tools via CLI flag
./slide-mcp-server --api-key YOUR_KEY --disabled-tools "slide_agents,slide_backups"

# Disable tools via environment variable
export SLIDE_DISABLED_TOOLS="slide_devices,slide_users"
./slide-mcp-server --api-key YOUR_KEY

# Combined with permission modes
./slide-mcp-server --api-key YOUR_KEY --tools reporting --disabled-tools "slide_snapshots"

# Whitespace is handled gracefully
./slide-mcp-server --api-key YOUR_KEY --disabled-tools " slide_agents , slide_backups , slide_devices "

Available Tool Names (v4.0.0)

slide_overview — inventory + health + per-client/per-device summaries
slide_files — file search + restore + push
slide_recovery — VMs + image exports + DR networks
slide_audit — account audit log
slide_clients — client CRUD
slide_admin — users + accounts
slide_agents, slide_devices, slide_snapshots, slide_backups, slide_alerts — lower-level CRUD with v4 task-oriented additions
list_all_clients_devices_and_agents — backward-compat alias for slide_overview operation=inventory

Key Features

Precedence: CLI flags take precedence over environment variables
Whitespace Handling: Extra spaces around tool names are automatically trimmed
Error Messages: Clear error messages when attempting to use disabled tools
Combined Filtering: Works alongside permission modes for layered access control
Transparency: Logs which tools are disabled on server startup

Use Cases

# Read-only server (only list/get/search/browse operations)
./slide-mcp-server --tools read-only --disabled-tools "slide_admin"

# Allow restores but hide the recovery tool entirely
./slide-mcp-server --tools safe --disabled-tools "slide_recovery"

# Monitoring-only setup
./slide-mcp-server --tools read-only --disabled-tools "slide_recovery,slide_files"

When a disabled tool is called, the server returns:

{
  "error": {
    "code": -32601,
    "message": "Tool 'slide_agents' is disabled"
  }
}

🔒 Permission Modes

v4.0.0 collapsed the permission system from four tiers down to three. Legacy mode names (reporting, restores, full-safe) are silently aliased so existing CLI flags / Claude Desktop user configs keep working.

Permission Levels

`read-only` — Pure read access

Use Case: monitoring dashboards, on-call sanity checks, anything that should never accidentally mutate state.

✅ Allowed: every list/get/search/browse op, every Resource read, every Prompt
❌ Blocked: every create/update/delete/start/restore/etc.
Legacy alias: reporting

./slide-mcp-server --api-key YOUR_KEY --tools read-only

`safe` — Default; everything except destructive ops

Use Case: general MSP / IT admin workflows including restores, agent settings, backup launches, alert resolution.

✅ Allowed: read-only + create/update for restores, agents, devices, networks, alerts, settings, backup launch
❌ Blocked: deletes (delete, delete_vm, delete_image, delete_network, delete_restore, delete_vlan, ...) and device power control (poweroff, reboot)
Legacy aliases: restores, full-safe

./slide-mcp-server --api-key YOUR_KEY --tools safe
# (this is the default; the flag is optional)

`full` — Complete access

Use Case: advanced admins who need to delete agents, delete snapshots, or remotely power-cycle a device.

✅ Allowed: every operation, including the irreversible ones
⚠️ Includes agent deletion, snapshot deletion, device poweroff/reboot

./slide-mcp-server --api-key YOUR_KEY --tools full

Permission Matrix

Operation category	`read-only`	`safe`	`full`
List / get / search / browse	✅	✅	✅
File search + restore + push	❌	✅	✅
Boot recovery VMs / export images	❌	✅	✅
DR network create / update	❌	✅	✅
Agent / device settings updates	❌	✅	✅
Alert resolution	❌	✅	✅
Backup launch	❌	✅	✅
Deletes (agents, snapshots, VMs, networks, port-forwards, ...)	❌	❌	✅
Device poweroff / reboot	❌	❌	✅

Security Recommendations

Production monitoring: read-only
Day-to-day MSP admin: safe (default)
Cleanup / decommissioning workflows: full

Usage with VS Code

For VS Code integration, add the following JSON block to your User Settings (JSON) file. You can do this by pressing Ctrl + Shift + P and typing Preferences: Open User Settings (JSON).

Optionally, you can add it to a file called .vscode/mcp.json in your workspace. This will allow you to share the configuration with others.

Note that the mcp key is not needed in the .vscode/mcp.json file.

{
  "mcp": {
    "inputs": [
      {
        "type": "promptString",
        "id": "slide_api_key",
        "description": "Slide API Key",
        "password": true
      }
    ],
    "servers": {
      "slide": {
        "command": "/path/to/slide-mcp-server",
        "env": {
          "SLIDE_API_KEY": "${input:slide_api_key}"
        }
      }
    }
  }
}

With custom configuration and disabled tools:

{
  "mcp": {
    "inputs": [
      {
        "type": "promptString",
        "id": "slide_api_key",
        "description": "Slide API Key",
        "password": true
      }
    ],
    "servers": {
      "slide": {
        "command": "/path/to/slide-mcp-server",
        "env": {
          "SLIDE_API_KEY": "${input:slide_api_key}",
          "SLIDE_TOOLS": "read-only",
          "SLIDE_DISABLED_TOOLS": "slide_admin"
        }
      }
    }
  }
}

💡 Usage Examples

Health summary

{
  "name": "slide_overview",
  "arguments": {"operation": "health"}
}

"Did backups run last night for ACME?"

{
  "name": "slide_backups",
  "arguments": {
    "operation": "status_for_client",
    "client_id": "c_0123456789ab",
    "hours": 24
  }
}

Find a file across every snapshot for an agent

{
  "name": "slide_files",
  "arguments": {
    "operation": "search",
    "agent_id": "a_0123456789ab",
    "search_term": "Q4-budget"
  }
}

Find every snapshot that contains a specific path

{
  "name": "slide_files",
  "arguments": {
    "operation": "versions",
    "agent_id": "a_0123456789ab",
    "path": "C:\\Users\\bob\\Documents\\Q4-budget.xlsx"
  }
}

Boot a recovery VM

{
  "name": "slide_recovery",
  "arguments": {
    "operation": "boot_vm",
    "snapshot_id": "s_0123456789ab",
    "device_id": "d_0123456789ab",
    "cpu_count": 4,
    "memory_in_mb": 8192,
    "network_type": "network-id",
    "network_source": "n_0123456789ab"
  }
}

Generate an RDP bookmark for a running VM

{
  "name": "slide_recovery",
  "arguments": {
    "operation": "get_rdp_bookmark",
    "virt_id": "v_0123456789ab"
  }
}

Audit log: what changed in the last 24 hours?

{
  "name": "slide_audit",
  "arguments": {"operation": "recent", "hours": 24}
}

Triage open alerts (sorted worst-first)

{
  "name": "slide_alerts",
  "arguments": {"operation": "triage"}
}

Start a backup

{
  "name": "slide_backups",
  "arguments": {"operation": "start", "agent_id": "a_0123456789ab"}
}

Export a snapshot as a VHDX image

{
  "name": "slide_recovery",
  "arguments": {
    "operation": "export_image",
    "snapshot_id": "s_0123456789ab",
    "device_id": "d_0123456789ab",
    "image_type": "vhdx"
  }
}

Build

# Build for current platform
make build

# Build for all platforms
make build-all

# Create release packages
make release

# View available commands
make help

Release Process

Unified Release Script (Recommended)

The release-all-in-one.sh script automates the entire release workflow:

# Auto-increment patch version and create release
./scripts/release-all-in-one.sh

# Release specific version
./scripts/release-all-in-one.sh v2.5.0

# Test release process without making changes
./scripts/release-all-in-one.sh --dry-run v2.5.0

# Build locally without pushing to GitHub
./scripts/release-all-in-one.sh --no-push

# Re-release existing version (overwrite tag and GitHub release)
./scripts/release-all-in-one.sh --force v2.5.0

What it does:

Checks prerequisites (Go, GitHub CLI, jq)
Validates git status (clean working directory, on main branch)
Updates version numbers in Makefile and config.go
Runs tests (optional with --skip-tests)
Commits version changes
Builds binaries for all platforms
Signs and notarizes macOS binaries (if credentials provided)
Creates release packages (.tar.gz for Unix, .zip for Windows)
Generates SHA256 checksums
Creates release notes from git history
Tags the release
Pushes to GitHub
Creates GitHub release with all assets

Environment Variables (REQUIRED for releases):

APPLE_ID: Apple ID for macOS notarization
APP_SPECIFIC_PASSWORD: App-specific password for notarization

⚠️ Important: macOS binaries MUST be signed and notarized to work on user systems. The script enforces this requirement and will fail if credentials are not provided. For local testing only, use --no-push --skip-macos-signing.

Setting Up Credentials (One-Time Setup):

# Option 1: Use credentials file (Recommended - saves credentials locally)
# File is already created at scripts/release-env.sh and gitignored for security
source scripts/release-env.sh

# Option 2: Set manually each time
export APPLE_ID='your-apple-id@example.com'
export APP_SPECIFIC_PASSWORD='xxxx-xxxx-xxxx-xxxx'

Getting Apple Credentials:

Apple ID: Your Apple Developer account email
App-Specific Password:
- Go to appleid.apple.com
- Sign in and navigate to Security section
- Click "Generate Password" under App-Specific Passwords
- Copy the generated password (format: xxxx-xxxx-xxxx-xxxx)
Developer ID Certificate: Install from Apple Developer account
- Required for code signing
- Must be "Developer ID Application" certificate

Security Note: The scripts/release-env.sh file is gitignored and will never be committed to the repository.

Options:

--dry-run: Simulate without making changes
--skip-tests: Skip running tests
--no-push: Build locally without GitHub operations
--skip-macos-signing: Skip signing (TESTING ONLY, must use with --no-push)
--force or -f: Force overwrite existing release (deletes tag and GitHub release)
--help: Show detailed help

Manual Release (Legacy)

For manual control, use the existing automated-release.sh script (see scripts/RELEASE_README.md).

Architecture Benefits

For LLMs

Reduced Complexity: 14 meta-tools vs 52+ individual tools
Logical Grouping: Related operations organized together
Consistent Interface: All meta-tools follow the same operation pattern
Better Context: Less tool switching, more focused conversations

For Developers

Maintainable: Each meta-tool in its own file
Extensible: Easy to add new operations to existing categories
Backward Compatible: All original functionality preserved
Schema Validation: Conditional parameter validation per operation

License

This MCP server is licensed under the MIT License. This means you are free to use, modify, and distribute the software, subject to the terms and conditions of the MIT License. For more details, please see the LICENSE file in the project repository.

Name		Name	Last commit message	Last commit date
Latest commit History 105 Commits
docs		docs
dxt		dxt
release		release
scripts		scripts
test_scripts		test_scripts
.gitignore		.gitignore
AGENTS.md		AGENTS.md
CHANGELOG.md		CHANGELOG.md
Makefile		Makefile
README.md		README.md
annotations.go		annotations.go
api.go		api.go
api_audit.go		api_audit.go
api_files.go		api_files.go
api_v127.go		api_v127.go
base_handler.go		base_handler.go
config.go		config.go
debug.go		debug.go
doctor.go		doctor.go
format.go		format.go
go.mod		go.mod
go.sum		go.sum
hints.go		hints.go
main.go		main.go
mcp.go		mcp.go
name_resolver.go		name_resolver.go
output_schemas.go		output_schemas.go
overview_helpers.go		overview_helpers.go
prompts.go		prompts.go
registry.go		registry.go
resources.go		resources.go
server.go		server.go
server_test.go		server_test.go
slide-mcp-server		slide-mcp-server
smoke_test.sh		smoke_test.sh
snapshots_test.json		snapshots_test.json
tools_admin.go		tools_admin.go
tools_agents.go		tools_agents.go
tools_alerts.go		tools_alerts.go
tools_audit.go		tools_audit.go
tools_backups.go		tools_backups.go
tools_clients.go		tools_clients.go
tools_devices.go		tools_devices.go
tools_files.go		tools_files.go
tools_help.go		tools_help.go
tools_overview.go		tools_overview.go
tools_recovery.go		tools_recovery.go
tools_snapshots.go		tools_snapshots.go
tools_v127.go		tools_v127.go

Folders and files

Latest commit

History

Repository files navigation

Slide MCP Server

What to ask first

What's new in v5.0.0

What's new in v4.0.0

🚀 Go Binary Implementation ⚡

⚡ Install

🟢 Claude Desktop - drag and drop (recommended)

🟦 Claude Code

🟪 Cursor / other MCP hosts

🛠️ From source / on a new dev box

🔏 Code signing setup (release builds only)

🎯 Architecture

Example Usage Pattern

Available Meta-Tools (v5.0.0)

💡 Help & onboarding

🧭 Overview & inventory

🔎 Files (the headline v4 capability)

🛟 Recovery

📜 Audit (compliance / "what changed?")

🗂️ Snapshots, backups, alerts, agents, devices

👥 Clients & admin

🔁 Backward-compat shim

MCP Prompts (slash-command workflows)

MCP Resources (cheap pre-baked context)

Token-conscious response shape

name_hint: no IDs to memorise

Key Features

🔐 Infrastructure Management

💽 Data Protection & Recovery

☁️ Virtualization

📊 Monitoring & Organization

📋 Professional Data Presentation

📦 Detailed installation reference

Getting an API Key

Pre-built downloads

Build from source

Manual Claude Desktop config

Test Your Installation

🔧 CLI Arguments & Configuration

Command Line Arguments

Examples

🚫 Disabling Specific Tools

Usage Examples

Available Tool Names (v4.0.0)

Key Features

Use Cases

🔒 Permission Modes

Permission Levels

read-only — Pure read access

safe — Default; everything except destructive ops

full — Complete access

Permission Matrix

Security Recommendations

Usage with VS Code

💡 Usage Examples

Health summary

"Did backups run last night for ACME?"

Find a file across every snapshot for an agent

Find every snapshot that contains a specific path

Boot a recovery VM

Generate an RDP bookmark for a running VM

Audit log: what changed in the last 24 hours?

Triage open alerts (sorted worst-first)

Start a backup

Export a snapshot as a VHDX image

Build

Release Process

Unified Release Script (Recommended)

Manual Release (Legacy)

Architecture Benefits

For LLMs

For Developers

License

About

Resources

Uh oh!

`read-only` — Pure read access

`safe` — Default; everything except destructive ops

`full` — Complete access