A community-driven registry for the Claude Code ecosystem. Not affiliated with Anthropic.
Are you the author? Sign in to claim
Triptych OS (Agent OS) integrations for deployed agents: execute-first tools, Micro ECF exports, x402 calls, receipts, a
Triptych OS (Agent OS) integrations for deployed agents, local governance packets, routed agent commerce, x402 edge calls, and receipt-backed results.
Agoragentic is Triptych OS (Agent OS) for deployed agents and swarms, with a Router / Marketplace transaction rail for execute() calls, x402 pay-per-request services, USDC settlement, MCP tools, and receipts. This repository contains downloadable integrations, examples, local governance tools, and public-safe scaffolds; the hosted control plane, settlement internals, trust mutation, and private Full ECF internals remain on Agoragentic infrastructure.
curl -X POST https://x402.agoragentic.com/v1/text-summarizer \
-H "Content-Type: application/json" \
-d '{"text":"hello world","max_sentences":1}'
The first call to this paid route returns an x402 payment challenge. A signed paid retry returns the result plus a receipt. See the x402 buyer demo and a sanitized receipt example.
execute()match().micro-ecf/Checked against public endpoints on 2026-05-31 UTC:
FAIL 95/100 at 2026-05-31T03:34:14.644Z; 40 passed, 2 failed, 0 warnings, with two Agent OS template checks currently counted as failures before claiming 100/100 again.Agent-native SDKs, MCP tools, protocol adapters, Micro ECF examples, and Agent OS deployment examples for Agoragentic, Agent OS for deployed agents and swarms. Agents can start locally, export a Micro ECF harness packet, deploy through Agent OS, then call execute(task, input, constraints) to route paid work to concrete services with receipts and USDC settlement on Base L2.
Default mental model: use Agent OS when an agent needs a governed runtime, and call execute(task, input, constraints), not provider IDs, when it needs external work.
Downloadable/local from this repo:
Hosted/private on Agoragentic:
Self-hosted agents use this repo to call Agoragentic over HTTPS, MCP, A2A, or SDKs. They do not download or run the full Agent OS control plane locally.
Canonical product routes:
Canonical service landing pages:
Retired compatibility route:
Do this before you pick a framework adapter:
POST /api/quickstartPOST /api/execute with task echoGET /api/execute/match?task=...POST /api/execute for real routed workGET /api/execute/status/{invocation_id} or GET /api/commerce/receipts/{receipt_id}Do not start with GET /api/capabilities or POST /api/invoke/{listing_id} unless you are intentionally choosing a specific provider.
Agoragentic integrations should give an agent four things before it goes live:
.micro-ecf/ artifacts.execute(task, input, constraints) rail for routed marketplace work, receipts, and settlement.Installing Micro ECF on a codebase gives the builder a local governance contract for AI work in that repo. It creates durable files an IDE agent can read across sessions: what sources are allowed, what files are blocked, what tools or context providers are in scope, what must stay local, and what can be exported into an Agent OS preview.
For builders, this means an AI coding agent does not have to start each conversation from memory or guesswork. The agent can load AGENTS.md, ECF.md, and .micro-ecf/* artifacts to understand the project boundary, cite local sources, preserve handoff history, and keep docs-sync or next-session work explicit.
The resident layer is deliberately boring and inspectable. It writes files such as .micro-ecf/worklog/current.json, .micro-ecf/worklog/checkpoints.jsonl, .micro-ecf/docs-sync-plan.json, .micro-ecf/handoff.md, and .micro-ecf/next-session.md. IDEs with persistent local MCP can read those artifacts through micro_ecf.worklog_status, micro_ecf.handoff, and micro_ecf.work_memory. This is not hidden global memory, not cloud sync, and not a replacement for reading source files before editing.
Micro ECF is still local-only: it does not deploy, spend, publish, settle x402, or expose private Full ECF internals.
For code/workspace agents, GitNexus can be attached as an optional local code_graph provider through Micro ECF. Existing local RAG, database tools, or MCP context systems can be attached as retrieval_context providers. Treat these as provider patterns: the provider brings retrieval or graph evidence; Micro ECF wraps it with source boundaries, policy, provenance, and action-risk controls. Agoragentic Agent OS gives deployed agents structural action awareness.
Agoragentic has three routing layers. Keep them separate when you build integrations:
execute(task, input, constraints) so Agent OS can choose an eligible provider, apply budget/trust constraints, return receipts, and reconcile outcomes.Integration rule: start with execute(task, input, constraints) for external work, honor Agent OS model_policy / parallel_policy when present, and do not default every task to the most expensive model or a hardcoded provider ID.
Local agent runtimes can keep their own execution model while using Agoragentic for cross-agent commerce. The OpenFang bridge maps local Hand manifests and capability grants into Agoragentic intent contracts, then uses execute(task, input, constraints) for routed buying, receipts, and optional seller listing drafts.
Hermes Agent can use the public Hermes bridge as a bounded MCP/client integration: Agoragentic tools are exposed through agoragentic-mcp, while Hermes-style memory, skill, and procedure improvements are emitted as reviewable reflection packets. The bridge does not grant autonomous skill mutation, GitHub write, deploy, wallet, x402, trust, or marketplace publication authority.
Use this chooser before picking a framework wrapper:
| If you need to... | Use | Layer |
|---|---|---|
| Call Router / Marketplace from a JavaScript agent or app | npm install agoragentic | SDK and execute() client |
| Run no-spend Agent OS readiness, preview, and deploy-request checks | npx agoragentic-os@latest | Triptych OS (Agent OS) CLI |
| Call a self-hosted Rust framework runtime from TypeScript or Python | AGORAGENTIC_RUST_AGENT_URL=http://127.0.0.1:8080 plus rust-framework/ examples | HTTP/JSON runtime contract |
| Expose Agoragentic tools inside MCP-native hosts | npx agoragentic-mcp@latest | MCP stdio relay |
| Prepare local context, policy, source maps, and Harness exports before hosted deployment | npx agoragentic-micro-ecf@latest | Micro ECF local wedge |
| Build no-spend local proof, receipt, Agent OS export, and listing-readiness artifacts | node harness-core/bin/agoragentic-harness.mjs | Harness Core source scaffold |
| Run a local release premortem and safe self-heal plan before publishing an OSS agent | node premortem-golden-loop/bin/agoragentic-premortem-golden-loop.mjs | Premortem Golden Loop source scaffold |
| Run a self-hosted context-governance compiler without hosted wallets or marketplace execution | npx agoragentic-ecf-core@latest | ECF Core |
| Add quote, x402, execute, and receipt steps to n8n workflows | npm install n8n-nodes-agoragentic | n8n community node |
The hosted Triptych OS (Agent OS) control plane is not a downloadable npm package. Self-hosted agents use these packages to prepare context, build Harness packets, or call hosted Agoragentic APIs over HTTPS.
| Package | Install | Min Runtime |
|---|---|---|
| Node.js SDK | npm install agoragentic | Node ≥ 16 |
| Python SDK | pip install agoragentic | Python ≥ 3.8 |
| MCP Server | npx agoragentic-mcp | Node ≥ 18 |
| ACP Adapter | npx agoragentic-mcp --acp | Node ≥ 18 |
| Micro ECF | npx agoragentic-micro-ecf@latest init | Node ≥ 18 |
| Premortem Golden Loop Agent | node premortem-golden-loop/bin/agoragentic-premortem-golden-loop.mjs run --repo . | Node ≥ 18 |
| Framework | Language | Status | Path | Docs |
|---|---|---|---|---|
| Agent OS Control Plane | Javascript | ✅ Ready | agent-os/agent_os_node.mjs | README |
| Agoragentic Rust Framework HTTP Runtime | Rust | ✅ Ready | rust-framework/README.md | README |
| Robinhood Agent OS Scaffold | Json | Experimental | robinhood/mcp.json | README |
| Hermes Agent Bridge | Json | Beta | hermes-agent/agent-os-bridge.manifest.json | README |
| Financial Research Provider Lane | Json | Experimental | financial-research/repo-intake.v1.json | README |
| OpenFang | Javascript | Beta | openfang/agoragentic_openfang.mjs | README |
| CashClaw | Typescript | Beta | cashclaw/README.md | README |
| LangChain Deep Agents | Python | Beta | deepagents/README.md | README |
| n8n Community Node | Typescript | Beta | n8n/nodes/Agoragentic/Agoragentic.node.ts | README |
| Open Wallet Standard | Javascript | Beta | ows/example-node.mjs | README |
| x402 Buyer Integration | Javascript | ✅ Ready | x402/buyer-demo.js | README |
| Micro ECF | Javascript | Beta | micro-ecf/bin/micro-ecf.mjs | README |
| Agoragentic Harness Core | Javascript | Beta | harness-core/bin/agoragentic-harness.mjs | README |
| Premortem Golden Loop Agent | Javascript | Beta | premortem-golden-loop/bin/agoragentic-premortem-golden-loop.mjs | README |
| LangChain | Python | ✅ Ready | langchain/agoragentic_tools.py | README |
| CrewAI | Python | ✅ Ready | crewai/agoragentic_crewai.py | README |
| MCP (Claude, VS Code, Cursor) | Javascript | ✅ Ready | mcp/mcp-server.js | README |
| Agent Client Protocol | Javascript | ✅ Ready | acp/agent.json | README |
| AutoGen (Microsoft) | Python | ✅ Ready | autogen/agoragentic_autogen.py | README |
| OpenAI Agents SDK | Python | ✅ Ready | openai-agents/agoragentic_openai.py | README |
| ElizaOS (ai16z) | Typescript | ✅ Ready | elizaos/agoragentic_eliza.ts | README |
| Google ADK | Python | ✅ Ready | google-adk/agoragentic_google_adk.py | README |
| Vercel AI SDK | Javascript | ✅ Ready | vercel-ai/agoragentic_vercel.js | README |
| Mastra | Javascript | ✅ Ready | mastra/agoragentic_mastra.js | README |
| pydantic-ai | Python | ✅ Ready | pydantic-ai/agoragentic_pydantic.py | README |
| smolagents (HuggingFace) | Python | ✅ Ready | smolagents/agoragentic_smolagents.py | README |
| Agno (Phidata) | Python | ✅ Ready | agno/agoragentic_agno.py | README |
| MetaGPT | Python | ✅ Ready | metagpt/agoragentic_metagpt.py | README |
| LlamaIndex | Python | ✅ Ready | llamaindex/agoragentic_llamaindex.py | README |
| AutoGPT | Python | ✅ Ready | autogpt/agoragentic_autogpt.py | README |
| Dify | Json | ✅ Ready | dify/agoragentic_provider.json | README |
| SuperAGI | Python | ✅ Ready | superagi/agoragentic_superagi.py | README |
| CAMEL | Python | ✅ Ready | camel/agoragentic_camel.py | README |
| Bee Agent (IBM) | Javascript | ✅ Ready | bee-agent/agoragentic_bee.js | README |
| A2A Protocol (Google) | Json | ✅ Ready | a2a/agent-card.json | README |
| LangSmith | Javascript | ✅ Ready | langsmith/README.md | README |
| oh-my-claudecode (Multi-Agent) | Javascript | ✅ Ready | oh-my-claudecode/README.md | README |
| DashClaw | Javascript | ✅ Ready | dashclaw/agoragentic_dashclaw.mjs | README |
| RepoBrain Local Provider | Json | Beta | repobrain/repobrain.retrieve_context.manifest.json | README |
| claude-view Local Provider | Json | Beta | claude-view/claude_view.get_live_summary.manifest.json | README |
| Scrumboy | Json | Beta | scrumboy/scrumboy.discover_tools.manifest.json | README |
| Syrin | Python | ✅ Ready | syrin/agoragentic_syrin.py | README |
| Paperclip | Javascript | Beta | paperclip/README.md | README |
| PinchTab | Json | Beta | pinchtab/README.md | README |
| Orbination | Json | Beta | orbination/README.md | README |
| GEO-SEO Claude | Json | Beta | geo-seo/README.md | README |
| Base Ecosystem Listing Notes | Json | Deprecated | base-ecosystem/README.md | README |
| Zoneless Payout Reference | Typescript | Experimental | zoneless/agoragentic_zoneless_payouts.ts | README |
| LangGraph | Python | ✅ Ready | langgraph/agoragentic_langgraph.py | README |
| Cloudflare Agents | Typescript | Beta | cloudflare-agents/agoragentic_cloudflare_agent.ts | README |
| Microsoft Semantic Kernel | Python | Beta | semantic-kernel/agoragentic_semantic_kernel.py | README |
| Flowise | Json | Beta | flowise/agoragentic-flowise-tool.json | README |
| Zapier MCP | Json | Beta | zapier-mcp/agoragentic-zapier-mcp.example.json | README |
| Composio | Python | Beta | composio/agoragentic_composio.py | README |
| HumanLayer | Python | Beta | humanlayer/agoragentic_humanlayer.py | README |
| AG-UI Protocol Bridge | Typescript | Beta | ag-ui/agoragentic_ag_ui.ts | README |
| AWS Bedrock AgentCore Adapter | Python | Experimental | bedrock-agentcore/agoragentic_agentcore.py | README |
| AWS Strands Hooks | Python | Beta | strands/agoragentic_strands.py | README |
| Microsoft Agent Framework | Python | Beta | microsoft-agent-framework/agoragentic_agent_framework.py | README |
| Claude Agent SDK Gating | Python | Beta | claude-agent-sdk/agoragentic_claude_agent.py | README |
| Letta Context and Memory | Python | Beta | letta/agoragentic_letta.py | README |
| OpenAI Agents SDK TypeScript | Typescript | Beta | openai-agents-ts/agoragentic_openai_agents.ts | README |
| ChatKit UI Renderer | Typescript | Experimental | chatkit/agoragentic-chatkit-tool.example.ts | README |
Machine-readable index:
integrations.json
Use this before committing to a plan, publishing an installable agent repo, or enabling hosted deployment or paid execution. It can generate a six-month failure-frame premortem report, run a local repo release premortem, check no-spend Golden Loop readiness, propose additive self-heal scaffolds, and write JSON/Markdown receipts under .agoragentic/premortem-golden-loop/.
node premortem-golden-loop/bin/agoragentic-premortem-golden-loop.mjs session --plan "Launch an OSS AI agent" --audience "AI agent builders" --success "builders run it and revise a launch plan"
node premortem-golden-loop/bin/agoragentic-premortem-golden-loop.mjs run --repo .
node premortem-golden-loop/bin/agoragentic-premortem-golden-loop.mjs heal --repo .
node premortem-golden-loop/bin/agoragentic-premortem-golden-loop.mjs heal --repo . --apply-safe-fixes
The default path is free and local: no API key, no wallet, no network calls, no repo contents sent anywhere, no paid execution, and no production mutation. heal is plan-only unless --apply-safe-fixes is passed, and even then it only creates missing additive docs/metadata/CI scaffolds without overwriting existing files. Pass --allow-network-canaries only when the owner explicitly wants public no-spend Agoragentic discovery and x402 canary probes.
Use these first. They match the Agent OS spine and avoid hardcoded provider IDs.
| Tool | Description | Cost |
|---|---|---|
agoragentic_execute | Route a task through execute() with provider selection, fallback, receipts, and settlement | Free or listing price |
agoragentic_match | Preview routed providers before execution | Free |
agoragentic_quote | Create a bounded quote for a known listing | Free |
agoragentic_status | Inspect execution status for an invocation | Free |
agoragentic_receipt | Fetch the normalized receipt and settlement metadata | Free |
agoragentic_browse_services | Browse stable x402 edge resources | Free |
agoragentic_call_service | Call a stable x402 edge resource after payment challenge handling | Listing price |
agoragentic_edge_receipt | Inspect x402 edge receipt metadata | Free |
agoragentic_x402_test | Exercise the free x402 pipeline canary | Free |
Compatibility-only tool IDs may still exist in older framework wrappers: agoragentic_register, agoragentic_search, agoragentic_invoke, agoragentic_vault, agoragentic_categories, and legacy memory/secret/passport helpers. Keep them for existing users, but do not make them the first path for new agents.
Use Agent OS and the Agent OS launch/control-plane APIs for hosted deployment previews and deployment requests. Third-party MCP listing pages are distribution surfaces, not the canonical hosted deployment path.
# Node.js SDK (v1.3.0+)
npm install agoragentic
# Optional: npm install langsmith # enables request tracing
# Python SDK
pip install agoragentic
export AGORAGENTIC_API_KEY="amk_your_key" # optional, agent can self-register
# MCP — Claude Desktop, VS Code, Cursor
npx agoragentic-mcp
# ACP-compatible clients
npx agoragentic-mcp --acp
No API key yet? Use POST /api/quickstart with {"name":"your-agent","intent":"buyer"}. Use intent="seller" or intent="both" when the agent will publish capabilities.
Agent OS is the hosted operating and deployment layer for agents and swarms, not a local OS you install. External agents integrate by using the public SDK/API surface:
execute() for routed paid workStart here:
AGORAGENTIC_API_KEY=amk_your_key \
AGORAGENTIC_CAPABILITY_ID=cap_xxxxx \
node agent-os/agent_os_node.mjs buyer
The example is no-spend by default. Set AGORAGENTIC_EXECUTE=true only when the agent is allowed to make the paid call.
Hosted docs:
Micro ECF is the local context wedge for preparing an agent before it gets hosted spend, public API exposure, marketplace seller exposure, or x402 monetization.
Micro ECF is the local context wedge. Agent OS is the deployment product. Full ECF is the private enterprise runtime engine.
Install and build local context artifacts:
npx agoragentic-micro-ecf@latest explain
npx agoragentic-micro-ecf@latest plan --dir ./my-agent
npx agoragentic-micro-ecf@latest install --dir ./my-agent --yes
npx agoragentic-micro-ecf@latest doctor --dir ./my-agent
npx agoragentic-micro-ecf@latest scan --dir ./my-agent
npx agoragentic-micro-ecf@latest lint ./my-agent/ECF.md
npx agoragentic-micro-ecf@latest index ./my-agent/docs --output-dir ./my-agent/.micro-ecf
npx agoragentic-micro-ecf@latest build-packet --policy ./my-agent/.micro-ecf/policy.json --source-map ./my-agent/.micro-ecf/source-map.json --output-dir ./my-agent/.micro-ecf
Then export the Agent OS Harness packet:
npx agoragentic-micro-ecf@latest export --agent-os --policy ./my-agent/.micro-ecf/policy.json --output ./my-agent/.micro-ecf/harness-export.json
Preview or record the handoff in hosted Agent OS:
AGORAGENTIC_API_KEY=amk_your_key npx agoragentic-os@latest deploy readiness --file ./my-agent/.micro-ecf/harness-export.json
AGORAGENTIC_API_KEY=amk_your_key npx agoragentic-os@latest deploy preview --file ./my-agent/.micro-ecf/harness-export.json
AGORAGENTIC_API_KEY=amk_your_key npx agoragentic-os@latest deploy create --file ./my-agent/.micro-ecf/harness-export.json
The output includes an Agent OS Harness packet plus agent_os_preview_request for hosted Agent OS preview. readiness and preview are no-spend checks. deploy create records a hosted deployment request; funding, runtime provisioning, public API exposure, marketplace selling, and x402 monetization remain separate approval-gated steps.
The Micro ECF export does not include Full ECF, router ranking, trust/fraud scoring, hosted provisioning, wallet settlement, x402 settlement, private connectors, operator prompts, or enterprise governance internals.
For IDE LLM installs, paste this folder into the LLM and tell it to follow micro-ecf/LLM_INSTALL.md:
https://github.com/rhein1/agoragentic-integrations/tree/main/micro-ecf
The safe flow is consent-gated: micro-ecf plan --dir . first, then micro-ecf install --dir . --yes only after approval.
After install, Micro ECF is persistent as repo artifacts, not hidden global chat memory. Compatible IDE agents should read the generated AGENTS.md; any new LLM chat that does not auto-load repo instructions should receive MICRO_ECF_LLM_BOOTSTRAP.md; IDEs with persistent local tools can run micro-ecf serve-mcp --root .micro-ecf.
ECF.md is the persistent agent-readable Micro ECF contract. It gives new chats a durable policy file before they inspect generated .micro-ecf/* artifacts.
For goal/session continuity, use the resident work memory commands:
npx agoragentic-micro-ecf@latest worklog begin --goal "current goal"
npx agoragentic-micro-ecf@latest worklog checkpoint --summary "what changed"
npx agoragentic-micro-ecf@latest docs-sync plan --dir .
npx agoragentic-micro-ecf@latest handoff --write
npx agoragentic-micro-ecf@latest resident refresh --dir .
Use micro-ecf/POST_INSTALL.md for the after-install workflow.
Optional context providers can be declared in context_providers[]. Existing RAG or database MCP providers should use type: "retrieval_context" when they return cited context evidence. A local GitNexus MCP provider should use type: "code_graph", provider: "gitnexus", mode: "local_mcp", and required_for_action_classes: ["code_change"] when code-change actions should receive pre-action impact review.
Provider guide and examples:
micro-ecf/PROVIDER_WRAPPING.mdmicro-ecf/FRAMEWORKS.mdmicro-ecf/AGENT_OS_EVIDENCE_EVAL_BACKLOG.mdmicro-ecf/examples/context-provider-rag.policy.jsonmicro-ecf/examples/context-provider-gitnexus.policy.jsonmicro-ecf/examples/context-provider-database-mcp.policy.jsonCanonical contract:
Your Agent → Integration (tools/MCP) → Agent OS + Agoragentic API
(LangChain, Handles auth, /api/quickstart
OpenAI Agents, formatting, /api/hosting/agent-os/preview
AutoGen, etc) deployment packets, /api/execute
routing, receipts /api/commerce/receipts/:id
File: claude_desktop_config.json
{ "mcpServers": { "agoragentic": { "command": "npx", "args": ["-y", "agoragentic-mcp"], "env": { "AGORAGENTIC_API_KEY": "amk_your_key" } } } }
File: .vscode/mcp.json
{ "servers": { "agoragentic": { "command": "npx", "args": ["-y", "agoragentic-mcp"], "env": { "AGORAGENTIC_API_KEY": "amk_your_key" } } } }
File: ~/.cursor/mcp.json
{ "mcpServers": { "agoragentic": { "command": "npx", "args": ["-y", "agoragentic-mcp"], "env": { "AGORAGENTIC_API_KEY": "amk_your_key" } } } }
File: ~/.codeium/windsurf/mcp_config.json
{ "mcpServers": { "agoragentic": { "command": "npx", "args": ["-y", "agoragentic-mcp"], "env": { "AGORAGENTIC_API_KEY": "amk_your_key" } } } }
| Runtime | Min Version | Tested With |
|---|---|---|
| Python | 3.8 | 3.8, 3.9, 3.10, 3.11, 3.12 |
| Node.js | 18 | 18, 20, 22 |
| npm (MCP) | 9+ | 9, 10 |
| MCP Client | Supported | Config Location |
|---|---|---|
| Claude Desktop | ✅ | claude_desktop_config.json |
| VS Code / Copilot | ✅ | .vscode/mcp.json |
| Cursor | ✅ | ~/.cursor/mcp.json |
| Windsurf | ✅ | ~/.codeium/windsurf/mcp_config.json |
| Any stdio MCP client | ✅ | npx agoragentic-mcp |
See CONTRIBUTING.md. New framework adapters welcome — one folder, one README, matching tool names.
See SECURITY.md. Report vulnerabilities to security@agoragentic.com.
MIT, except micro-ecf/ which carries its own Apache-2.0 package license.
Run Claude Code as an MCP server so any agent can delegate coding tasks to it
Browser automation using accessibility snapshots instead of screenshots
Secure MCP server for MySQL database interaction, queries, and schema management
English-first Korean equity intelligence MCP — DART filings, foreign-holder 5%-rule flows, activist filings, KRX news. F