Atlas / MCP servers / swarmclawai / SwarmClaw

SwarmClawBLOCK

Name: SwarmClaw
Author: swarmclawai

mcp/swarmclawai/swarmclaw

MCP server by swarmclawai · agent framework · agent memory · agent runtime · Audited 2026-09-28

Open-source self-hosted AI agent runtime and multi-agent framework for autonomous agent swarms. Agent memory, MCP tools, schedules, delegation, and 23+ LLM providers (Claude, GPT, Gemini, OpenRouter, Ollama). A practical Claude Code and LangChain alternative.

Verdict

BLOCK

Grade

Trust score

66 /100

Exposed tools

190 157r · 31w · 2d

Transport

stdio · streamable-http

License

MIT

Stars

685

Overview

From the repository's own README, as read at the audited commit. Badges and raw HTML are left out.

[](https://github.com/swarmclawai/swarmclaw/actions/workflows/ci.yml) [](https://github.com/swarmclawai/swarmclaw/releases) [](https://www.npmjs.com/package/@swarmclawai/swarmclaw)

The self-hosted AI agent runtime and multi-agent framework for autonomous agents. Open-source agent swarms with durable agent memory, MCP tools, skills, delegation, schedules, and 23+ LLM providers — a practical Claude Code and LangChain alternative.

SwarmClaw is an open-source, self-hosted AI agent runtime and multi-agent framework. Run autonomous AI agents, agent swarms, and orchestrators with heartbeats, schedules, delegation, agent memory, runtime skills, and reviewed conversation-to-skill learning — across OpenClaw gateways, Claude, GPT, Gemini, OpenRouter, Ollama, and 23+ other providers. Use it as your AI agent dashboard, agent orchestration platform, and home base for self-hosted multi-agent AI workflows.

GitHub: https://github.com/swarmclawai/swarmclaw Docs: https://swarmclaw.ai/docs Website: https://swarmclaw.ai Discord: https://discord.gg/sbEavS8cPV Extension tutorial: https://swarmclaw.ai/docs/extension-tutorial

Screenshots

Read from source at commit 957c9e3d8e01OBSERVED · 2026-09-28

Connect

Built from this server's own package name, version and transport as found in its source — not copied from anyone's documentation, so it cannot drift against a page we do not control. Replace the environment placeholders with a token scoped to the least it needs.

claude-code

claude mcp add swarmclaw --env ACCESS_KEY=${ACCESS_KEY} --env COPILOT_GITHUB_TOKEN=${COPILOT_GITHUB_TOKEN} --env CREDENTIAL_SECRET=${CREDENTIAL_SECRET} --env ELEVENLABS_API_KEY=${ELEVENLABS_API_KEY} -- npx -y @swarmclawai/[email protected]

claude-desktop

{
  "mcpServers": {
    "swarmclaw": {
      "command": "npx",
      "args": [
        "-y",
        "@swarmclawai/[email protected]"
      ],
      "env": {
        "ACCESS_KEY": "${ACCESS_KEY}",
        "COPILOT_GITHUB_TOKEN": "${COPILOT_GITHUB_TOKEN}",
        "CREDENTIAL_SECRET": "${CREDENTIAL_SECRET}",
        "ELEVENLABS_API_KEY": "${ELEVENLABS_API_KEY}"
      }
    }
  }
}

Exposed tools (190)

157 read · 31 write · 2 destructive. Blast radius: 2 tools can delete or overwrite — an agent that can be talked into calling a tool can be talked into calling this one.

Tool	Risk	Description
`Alpha`	read	Test agent
`Analyst`	read
`Anthropic`	read	Claude models — strong for coding, analysis, and long-form reasoning.
`Assistant`	read
`Atlas`	read	A helpful GPT-powered assistant.
`Backlog`	read	Separate workspace
`Beta`	read	Second test agent
`Bolt`	read	A low-latency assistant powered by Groq.
`Builder`	read	Implements ideas, ships changes, and drives technical execution.
`Claude`	read	A helpful Claude-powered assistant.
`CodeBot`	read
`Coordinator`	read
`Data-Driven`	read	Always backs claims with numbers.
`DeepInfra`	read	Fast serverless inference for popular open-source models.
`DeepSeek`	read	High-value reasoning and coding models from DeepSeek.
`Default`	read	Default workspace
`Design`	read	Handles design critique and UI polish
`Editor`	read	Improves structure, tone, and quality before publishing.
`Gardener`	read	Nurtures ideas and lets them grow.
`Gemini`	read	A helpful Gemini-powered assistant.
`Grok`	read	A helpful assistant powered by xAI Grok.
`Groq`	read	Very fast inference with open and reasoning model options.
`Hal`	read
`Hal2k`	read	Heartbeat hygiene test
`Hermes`	read	A runtime-backed assistant powered by Hermes Agent.
`Local`	read	A local assistant running through Ollama.
`Maker`	read	Executes focused work items assigned by the user or other agents.
`Mistral`	read	A helpful assistant powered by Mistral.
`Molly`	read
`Mosaic`	read	A helpful assistant powered by Together AI.
`Nebius`	read	Wide catalog of 60+ open-source models via Nebius Token Factory.
`Ollama`	write	Run local open-source models or connect to Ollama Cloud.
`OpenAI`	read	Great default for most users. Fast, reliable GPT models.
`OpenClaw`	write	Deploy or connect official-only local and remote OpenClaw gateways, then map starter agents across your swarm by role, tag, or use case.
`OpenRouter`	read	One API key for a broad multi-provider model catalog through an OpenAI-compatible API.
`Operator`	read	Coordinates tasks, delegates work, and keeps the workspace moving.
`Ops`	read	Handles deploys and infrastructure
`Probe`	read	Heartbeat probe
`Prototyper`	read	Builds first, specs later.
`Requesty`	read	One API key for a broad multi-provider model catalog through an OpenAI-compatible gateway, with caching, failover, and cost controls.
`Researcher`	read	Collects facts, compares options, and produces structured findings.
`Reviewer`	read	Reviews plans and outputs for bugs, regressions, and quality gaps.
`Rex`	read	A helpful agent
`Room`	read
`Router`	read	A helpful assistant powered through OpenRouter.
`Sidekick`	read	A versatile assistant for everyday work, planning, and follow-through.
`Spark`	read	A helpful assistant powered by Fireworks AI.
`SwarmClaw`	read	AI agent orchestration
`Swarmy`	read	Helpful coding agent
`Test`	read
`TokenMix`	read	One OpenAI-compatible API relay for Claude, OpenAI, Gemini, DeepSeek, Qwen, and other hosted models.
`Triager`	read	Triages inbound messages into urgent, reply-needed, and informational buckets.
`Worker`	read
`Writer`	read
`a`	read	Tool A
`a2a`	read	A2A Protocol gateway
`activity`	read	Query activity feed events
`agents`	read	Manage agents
`approvals`	read	List and resolve human-loop approvals
`artifacts`	read	Resolve evidence artifacts for runs, missions, and tasks
`auth`	read	Access key auth helpers
`autonomy`	read	Inspect supervisor incidents and reflection output
`b`	read	Tool B
`browser`	read	Control a persistent browser profile. Supports low-level actions plus higher-level workflows like read_page, extract_links, extract_form_fields, extract_table, fill_form, submit_form, scroll_until, download_file, complete_web_task, profile, and reset_profile.
`c`	read	Tool C
`chatroom-policy`	read	Configure chatroom delegation refusal policies
`chatrooms`	read	Manage multi-agent chatrooms
`chats`	read	Manage agent chats and runtime controls
`claude-skills`	read	Read local Claude skills directory metadata
`clawhub`	write	Browse and install ClawHub skills
`config-versions`	read	Inspect and restore configuration version history
`connector_message_tool`	write	Send and manage outbound messages across chat platforms.
`connectors`	read	Manage chat connectors
`context_status`	read	Check token usage and context window limits.
`context_summarize`	read	Compact conversation history to free up space.
`cost-attribution`	read	Aggregate LLM cost by billing-code tags
`credentials`	read	Manage encrypted provider credentials
`custom_media_sender`	write	Send rendered media somewhere special.
`cwd_check`	read
`daemon`	read	Control background daemon
`delegate`	read	Delegate to a specialized backend (Claude, Codex, OpenCode, Gemini, Copilot, Droid, Cursor, Qwen) for code tasks: writing files, refactoring, debugging, code generation, and multi-file edits. Supports background jobs with action=status\|list\|wait\|cancel.
`delegation-jobs`	read	Delegation job status
`deploy-repair-workflow`	write	Repair deploy regressions in a stable order.
`deploy-triage`	write	Triage deploy issues in a fixed order.
`deploy-verification`	write	Verify deploy blockers, config, and smoke checks.
`deploy-workflow-learned`	write	Agent-scoped deploy workflow.
`dirs`	read	Directory listing and native picker
`dispatch-helper.md`	read	Dispatch through manage_skills status.
`documents`	read	Manage documents
`echo`	read
`edit_file`	write	Surgically replace a specific string within a file.
`email`	write	Send an email or check email configuration status. For sending: provide to, subject, and body. Optionally include html for rich formatting.
`eval`	write	Run agent evaluation scenarios
`execute`	write	Execute a bash script. Supports curl, jq, awk, sed, grep, and 70+ Unix commands.
`extension_creator_tool`	destructive	Create, read, edit, delete, or get the spec for writing new SwarmClaw extensions. Always call get_spec first to learn the correct format.
`extensions`	read	Manage extensions and marketplace
`external-agents`	read	Manage external agent runtimes
`files`	read	Serve and manage local files
`gateways`	read	Manage named OpenClaw gateway profiles
`generate_image`	read	Generate an image from a text prompt. The image is saved and a download link is returned. Use descriptive, detailed prompts for best results.
`generic-notes`	read	Store notes.
`github`	read	GitHub operations
`github-sync`	write	Stored GitHub flow.
`github-sync.md`	write	Sync GitHub issues into tasks.
`goals`	read	Manage goal hierarchy
`google-workspace-helper`	read	Automate Google Workspace docs and sheets.
`google_workspace`	write	Run a Google Workspace CLI (
`greet`	read	Greets
`healthz`	read	Public liveness probe
`incident-triage`	write	Triage incidents in a stable order.
`ip`	read	Get local IP/port metadata
`knowledge`	read	Manage knowledge base entries
`learned-skills`	read	Inspect agent-scoped learned skills
`logs`	destructive	Read or clear app logs
`mailbox`	read	Work with email inboxes. Actions: status, list_messages, list_threads, search_messages, read_message, download_attachment, reply, wait_for_email.
`manage_capabilities`	read	Discover available built-in tools or external extensions, search marketplaces, or request access to a direct tool or extension id with action=
`manage_platform`	read	Unified fallback tool for managing SwarmClaw resources when a more specific
`manage_protocols`	read	Structured orchestration workflows. Actions: list_templates, create_run (title, participantAgentIds, templateId or steps), run_status (runId), list_runs, run_action (runId, runAction: start\|pause\|resume\|cancel), create_template (name, description, tags), run_events (runId).
`mcp-registry`	read	Browse the public SwarmDock MCP Registry
`mcp-servers`	read	Manage MCP server configurations
`meeting-notes`	read	Capture meeting notes and follow-ups.
`memories`	read	Alias of memory command group
`memory`	read	Manage memory entries
`memory-images`	read	Fetch stored memory image assets
`memory_get`	read	Read a specific memory entry by id or key after search, keeping context focused.
`memory_search`	read	Search durable long-term memory for prior work, decisions, dates, people, preferences, or todos from earlier conversations. Prefer this before broader history tools.
`memory_tool`	write	Advanced long-term memory system. Store and update canonical durable facts across conversations; store/update will merge matching memories and retire superseded variants. Search defaults to durable memories unless sources explicitly include archive or working.
`memory_update`	write	Update or correct an existing durable memory when new information supersedes the old value.
`missions`	read	Manage autonomous missions
`monitor_tool`	write	Create durable waits that resume the agent when conditions trigger. Actions: create_watch, wait_until, wait_for_http, wait_for_file, wait_for_task, wait_for_webhook, wait_for_page_change, list_watches, get_watch, cancel_watch. For sys_info/ping/log tailing, use shell instead.
`my_custom_tool`	read	Does something useful
`notifications`	read	Manage in-app notifications
`openclaw`	read	OpenClaw discovery, gateway control, and runtime APIs
`openclaw_browser`	read	Control a browser through the OpenClaw CLI.
`openclaw_nodes`	read	Interact with connected OpenClaw nodes/devices.
`openclaw_workspace`	read	Versioning tools for the OpenClaw workspace.
`operations`	read	Operator triage and readiness summaries
`other-agent-skill`	read	Should not leak.
`perf`	read	Inspect or control runtime perf tracing
`pin_context`	read	Pin a fact, failed approach, blocker, or discovery to working memory so it survives context compaction and flows to subagents.
`ping`	read
`platform`	read	Consolidated platform tool. Use dotted action names:
`portability`	write	Export and import agent configurations
`preview-server`	read	Manage preview dev servers
`projects`	read	Manage projects
`prompt-helper.md`	read	Guidance-only workflow.
`protocols`	read	Manage Structured Session runs and templates
`providers`	read	Manage providers and model overrides
`read_file`	read	Reads a file
`replicate`	write	Run AI models on Replicate. Actions: run (create and wait for prediction), get (check prediction status), cancel (stop a prediction), get_model (model details), search (find models), status (check config).
`request_tool_access`	read	Ask the user for access to an extension I don\
`runs`	write	Session run queue/history
`schedule_wake`	write	Schedule a wake event (reminder) for yourself in this chatroom.
`schedules`	read	Manage schedules
`search`	read	Global search across app resources
`secrets`	read	Manage reusable encrypted secrets
`send_file`	write	Send a file to the user in chat. Use the returned /api/uploads/... links exactly as provided.
`sessions_tool`	read	Manage sessions and check identity. Actions: identity (whoami), list, history, spawn, update.
`settings`	write	Read/update app settings
`setup`	read	Setup and provider validation helpers
`share`	read	Public share links for missions, skills, and sessions
`shell`	write	Execute commands and manage processes. Use for git, curl, long-lived servers, and other host CLI tools.
`skill-suggestions`	read	Review conversation-derived skill drafts
`skills`	read	Manage reusable skills
`souls`	read	Browse and manage soul library templates
`spawn_subagent`	read	Delegate tasks to other agents.
`swarmdock`	read	SwarmDock marketplace
`swarmfeed`	read	SwarmFeed social network
`system-status`	read	Lightweight system health summary
`t`	read	d
`tasks`	read	Manage task board items
`test-hub-skill`	read	A ClawHub test skill.
`tool1`	read	My cool tool
`tts`	read	Text-to-speech endpoint
`upload`	write	Upload raw file/blob
`uploads`	read	Manage uploaded artifacts
`usage`	read	Usage and cost summary
`version`	write	Version and update checks
`wallets`	read	Manage agent wallets
`web`	read	Unified web access tool. Actions: search (web search), fetch/extract (read URL content), crawl (bounded same-origin crawl), api (raw HTTP request with method/headers/body).
`web_crawl`	write	Crawl a small set of pages starting from one URL. Same-origin by default, bounded by maxPages and maxDepth.
`web_extract`	read	Extract readable content from a URL with title and source URL included.
`web_fetch`	read	Read a specific URL and return readable page text.
`web_search`	read	Search the web and return ranked results with URLs and snippets.
`webhooks`	write	Manage and trigger webhooks
`whatsapp-voice-fallback`	read	Use a fallback voice synthesis path for WhatsApp delivery.
`whatsapp-voice-fallback-v2`	read	Revised fallback.
`workflow-states`	read	Manage customizable task workflow states
`workspace-helper`	read	Automate workspace docs.
`workspaces`	read	Manage logical workspaces (multi-workspace scaffolding)

Trust audit

BLOCKgrade D · trust 66/100 Do not install this without reading the findings. The audit found something that could harm you or your machine.

Layer	What it checks	Result
L0	Provenance & inventory	FAIL
L1	Static analysis of the code	FAIL
L2	Instruction surface (what it tells the agent)	PASS
L3	Class-specific surface	WARN
L4	Behavioural (sandbox)	SKIPPED

What the source does

Filesystem: declared (10 observation(s))
Network: declared (17 observation(s))
Shell: declared (5 observation(s))
Dependencies: not all pinned
Secrets in source: none-found

Findings (25)

HIGHInventory / provenance · inv.suspicious_name · CWE-1104

src/lib/skill-save-payload.ts

skill-save-payload.ts

Why it matters. member named after an attack tool

Fix. remove or justify

HIGHFilesystem / path · fs.credential_store · CWE-22, CWE-59

src/components/openclaw/openclaw-deploy-panel.tsx:1184

placeholder="~/.ssh/id_ed25519"

Why it matters. touches a credential store

HIGHFilesystem / path · fs.credential_store · CWE-22, CWE-59

src/lib/server/session-tools/index.ts:204

'**/.ssh/id_*', '**/.env', '**/.env.local', '**/.gnupg/**',

Why it matters. touches a credential store

HIGHNetwork egress · net.tls_off · CWE-200, CWE-319

src/lib/server/connectors/email.ts:71

: { rejectUnauthorized: false, checkServerIdentity: () => undefined }

Why it matters. certificate verification is disabled

Fix. leave verification on

HIGHNetwork egress · net.tls_off · CWE-200, CWE-319

src/lib/server/session-tools/email.ts:107

const tlsSocket = tls.connect({ socket, host: cfg.host, rejectUnauthorized: false }, () => {

Why it matters. certificate verification is disabled

Fix. leave verification on

HIGHNetwork egress · net.tls_off · CWE-200, CWE-319

src/lib/server/session-tools/email.ts:153

socket = tls.connect({ ...connectOpts, rejectUnauthorized: false }, () => {

Why it matters. certificate verification is disabled

Fix. leave verification on

MEDIUMNetwork egress · net.beacon_words · CWE-200, CWE-319

src/lib/server/untrusted-content.ts:6

{ code: 'credential_theft', re: /\b(?:api key|token|password|secret|credential)s?\b[\s\S]{0,40}\b(?:send|share|reveal|print|dump|exfiltrat)/i, note: 'asks for secrets or credentials' },

MEDIUMNetwork egress · net.raw_ip · CWE-200, CWE-319

electron/main.ts:8

const DEV_URL_DEFAULT = 'http://127.0.0.1:3456'

MEDIUMNetwork egress · net.raw_ip · CWE-200, CWE-319

electron/server-lifecycle.ts:72

const url = `http://127.0.0.1:${port}`

MEDIUMNetwork egress · net.raw_ip · CWE-200, CWE-319

scripts/browser-e2e-smoke.ts:150

const baseUrl = `http://127.0.0.1:${port}`

MEDIUMNetwork egress · net.raw_ip · CWE-200, CWE-319

scripts/sandbox-browser-entrypoint.sh:58

if curl -sS --max-time 1 "http://127.0.0.1:${CHROME_CDP_PORT}/json/version" >/dev/null; then

MEDIUMObfuscation / stealth · obf.zero_width · CWE-506, CWE-94

src/components/chatrooms/reaction-picker.tsx:40

'🐶', '🐱', '🐭', '🐹', '🐰', '🦊', '🐻', '🐼', '🐻❄️', '🐨',

MEDIUMObfuscation / stealth · obf.zero_width · CWE-506, CWE-94

src/components/chatrooms/reaction-picker.tsx:121

'❤️🔥', '❤️🩹', '❣️', '💕', '💞', '💓', '💗', '💖', '💘', '💝',

MEDIUMObfuscation / stealth · obf.zero_width · CWE-506, CWE-94

src/components/chatrooms/reaction-picker.tsx:133

'🚩', '🎌', '🏴', '🏳️', '🏳️🌈', '🏳️⚧️', '🏴☠️', '🇺🇸', '🇬🇧', '🇯🇵',

MEDIUMFilesystem / path · mcp.destructive_tools · CWE-22, CWE-59

extension_creator_tool, logs

Why it matters. 2 tool(s) can delete or overwrite

Fix. prefer a read-only mode or scoped tokens; the page states the blast radius

LOWInventory / provenance · inv.hidden_file · CWE-1104

.eslint-baseline.json

.eslint-baseline.json

Why it matters. hidden member outside the usual dotfiles

Fix. review its purpose

LOWInventory / provenance · inv.suspicious_name · CWE-1104

src/lib/skill-save-payload.test.ts

skill-save-payload.test.ts

Why it matters. member named after an attack tool

Fix. remove or justify