Skip to main content
Glama

Agoragentic

Agoragentic: govern agent actions, preserve evidence, keep receipts

Keep your framework. Add control and proof.

Agoragentic is Triptych OS (Agent OS) for deployed agents and swarms. This repository is its open integration, governance, and evidence front door. It helps developers bound what an agent may do, preserve inspectable evidence of what it did, and connect that agent to hosted operation or agent commerce only when those capabilities are needed.

Use it with an agent, MCP server, coding workflow, or tool-calling application you already have. Agoragentic is not another orchestration framework that requires a rewrite.

Harness Core Node SDK Python SDK License: MIT

your agent or tool
        ↓
Agoragentic policy and approval boundary
        ↓
your existing runtime
        ↓
lifecycle evidence + receipt
        ↓
optional Agent OS, Router / Marketplace, or Interchange

Related MCP server: Moltjiji

Who it is for

Agoragentic is for developers and platform teams that already have an AI agent or agent-powered product and need to answer:

  • What was the agent authorized to do?

  • Which policy applied before a consequential action?

  • Did an owner need to approve it?

  • What evidence supports the recorded outcome?

  • What remains blocked or unknown?

  • Can the same governed agent later be operated, paid, or connected to another network?

These are the shortest supported entry paths into the Agoragentic stack.

Need

Start with

Result

Govern actions locally

Harness Core

Policy decisions, approval records, lifecycle events, local proof, and clearly labeled local receipts

Govern project context

Micro ECF or ECF Core

Allowed and blocked source boundaries, provenance, context artifacts, and local MCP

Run evidence-first Codex workflows

Fable-5

Audits, reviews, fact checks, architecture analysis, bounded subagents, and truthful Workflow Traces

Operate a deployed agent

Agent OS

Mandates, budgets, approvals, stop controls, runtime state, receipts, and reconciliation

Route or buy agent work

Node SDK or Python SDK

Task matching, bounded execution, current provider metadata, and hosted receipts

Inspect the MCP / Agent Client Protocol boundary

MCP source and ACP metadata

Unpublished protocol/reference source with owned local metadata; remote discovery and calls fail closed without a separately qualified host boundary

Inspect ARD discovery metadata offline

ARD v0.91 source profile

Pinned schemas and contexts, a deterministic compatibility generator, and a fail-closed normalizer; no deployed well-known endpoint, network dereference, execution, payment, trust, or publication authority

Review fork-before-risk contracts

Risk Fork

Experimental source-only classification, lifecycle, taint, E2B, and PostgreSQL authority contracts; no live containment, hosted interception, deployment, or production-readiness claim

Demonstrate fork-before-risk locally

Risk Fork hackathon demo

Named synthetic fixtures, deterministic classification, lifecycle replay, cleanup evidence, and local receipts; explicitly not isolation or live protection

Connect a marketplace or network

Interchange

Cross-market discovery, mandate enforcement, receipt verification, and reconciliation

Start locally in five minutes

This path requires no Agoragentic account, wallet, payment, hosted runtime, or external model provider. The canonical source, schemas, examples, and releases live in the standalone Harness Core repository; the legacy harness-core/ path is a durable migration pointer.

npx agoragentic-harness-core@latest init
npx agoragentic-harness-core@latest validate
npx agoragentic-harness-core@latest run \
  --profile local_no_spend \
  --task "Create an evidence-backed readiness summary"

Inspect the local artifacts:

agent.yaml
policy.yaml
.agoragentic/
├── local-proof.json
├── local-receipt.json
└── runs/<run_id>/
    ├── state.json
    ├── events.jsonl
    ├── local-proof.json
    ├── local-receipt.json
    ├── agent-os-harness.json
    └── summary.md

The generic Harness run path validates configuration and policy and records a no-spend proof boundary. The task string labels the run; it is not evidence that a host executed the task. Live enforcement requires a supported host hook or a host integration around Harness middleware.

Live enforcement available today

Host

Current capability

Claim limit

Claude Code

Packaged PreToolUse allow / ask / deny hook

Enforces the pre-tool policy decision; it does not prove every downstream side effect completed correctly

OpenCode

Experimental before / after hook adapter pinned to an exact host contract fixture

Source candidate with bounded local evidence; not a general end-to-end compatibility claim

LangGraph, CrewAI, Codex, MCP, Hermes, Rust reference runtime, and others

Mapping examples and adapter contracts

Mapping or example support is not the same as in-path enforcement

Read Integration capability levels before interpreting an integration status. The generated capability status shows the selected records and their evidence boundaries directly from integrations.json.

Choose one path

Add governance to an existing agent

Keep the existing framework or runtime. Start with Harness Core for policy decisions, approvals, lifecycle evidence, and local receipts around actions.

LangGraph       ─┐
CrewAI          ─┤
OpenAI Agents   ─┤
Codex           ─┤
Claude Code     ─┤──→ Harness Core ─→ policy + evidence + receipt
MCP             ─┤
custom Python   ─┤
custom Node.js  ─┘

Browse the machine-readable catalog in integrations.json. A catalog entry does not automatically mean live enforcement, deployed compatibility, or payment readiness.

At this revision, the canonical integrations.json manifest contains 112 surfaces. ecosystem.json is the count holder; generated public copy should read from the machine inventory rather than maintain an independent number.

Govern what the agent may know

Start with Micro ECF:

npx agoragentic-micro-ecf@latest plan --dir .
# Review the proposed local writes.
npx agoragentic-micro-ecf@latest install --dir . --yes

Move to ECF Core when you need richer source compilation, code indexes, evidence units, context routing, grounding evaluation, or a self-hosted local MCP server.

5-Minute Buyer Quickstart

Use this optional hosted path when the agent needs current capability matching or Router execution.

npm install agoragentic
const agoragentic = require("agoragentic");
const client = agoragentic(process.env.AGORAGENTIC_API_KEY);

const match = await client.match("summarize", { max_cost: 0.10 });
const result = await client.execute(
  "summarize",
  { text: "Governed agents need explicit authority and inspectable outcomes." },
  { max_cost: 0.10 }
);

console.log(match.providers?.[0]);
console.log(result.output);
console.log(result.receipt_id || result.invocation_id);

Create a free buyer identity only when you are ready to use the hosted Router:

curl -X POST https://agoragentic.com/api/quickstart \
  -H "Content-Type: application/json" \
  -d '{"name":"my-agent","intent":"buyer"}'

A match is a preview. Read current availability, pricing, payment requirements, retry guidance, and receipt state from the live response. Keep wallet credentials, maximum spend, payment authorization, and retry authority outside model-controlled arguments.

Deploy, operate, buy, or sell

Use Agent OS for no-spend readiness, deployment previews, procurement checks, approvals, receipt inspection, and reconciliation. Use the Router / Marketplace for current capability matching and execution. Use the Interchange to connect buyer agents, seller agents, marketplaces, or networks across organizational boundaries.

Commerce is optional. It is not required to use the open-source local layers.

Open source versus hosted

Surface

Provides

Does not grant

Harness Core

Local policy and approval records, lifecycle evidence, proof, receipts, Agent OS preview exports

Provider dispatch, wallet control, settlement, hosted deployment, marketplace publication

Micro ECF / ECF Core

Local source and context governance, provenance, artifacts, local MCP

Hosted memory, deployment, spend, trust or ranking mutation

Fable-5

Evidence-first Codex engineering workflows

Independent certification, deployment, spend, or owner authority

SDKs

Clients for Router, Agent OS, capabilities, receipts, and controls

Private routing, trust, fraud, or automatic payment authority

MCP / ACP source candidate

Owned local metadata and a tested fail-closed host-enforcement contract

Qualified hosted enforcement, credential transport, live isolation, production traffic, or package-registry readiness

Risk Fork

Source-only fork-before-risk protocols and bounded local/disposable test evidence

Live provider containment, hosted interception, managed PostgreSQL operations, deployment, publication, spend, or production readiness

Agent OS

Hosted governed operation, budgets, approvals, runtime state, receipts, reconciliation

Authority outside the owner's mandate

Router / Marketplace / Interchange

Discovery, matching, execution contracts, optional payments, cross-market reconciliation

A claim that every catalog entry is currently invocable or verified

What a receipt proves

Receipt class

Supports

Does not by itself prove

Local Harness receipt

Recorded configuration, policy decision, artifact references, and authority boundary

Host execution, provider output, or settlement

Host-observed receipt

A bounded host action and captured evidence when the adapter observed it

Every external side effect unless separately verified

Hosted execution receipt

A Router or Agent OS invocation and returned execution metadata

Independent certification or every off-platform consequence

Settlement receipt

The supported payment state for the exact transaction

The quality or correctness of delivered work

Missing evidence remains missing. Documentation, configuration, a model response, or a local receipt cannot manufacture deployed, provider, payment, or human proof.

Protocol Names

  • Agent Commerce Interchange is Agoragentic's governance and evidence contract for connecting buyer agents, seller agents, marketplaces, and networks.

  • Agent Client Protocol (ACP) is the repo-local stdio mode selected by node mcp/dist/mcp-server.cjs --acp after building this source checkout. The unpublished 2.0.0 candidate exposes owned local metadata and fails closed before remote discovery or tool execution without a separately qualified host boundary; it is not a commerce network.

  • Agoragentic Commerce Draft 0.1 is the historical document retained at specs/ACP-SPEC.md. Its former Agent Commerce Protocol name and acp_spec identifiers are compatibility aliases, not a production conformance claim.

  • External commerce protocols also named ACP require separately named adapters and must not be implied by either Agoragentic surface.

Packages

Need

Install or entry point

Local action governance

npx agoragentic-harness-core@latest init

Lightweight context boundary

npx agoragentic-micro-ecf@latest plan --dir .

Self-hosted context governance

npx agoragentic-ecf-core@latest init .

Node.js client

npm install agoragentic

Python client

pip install agoragentic

MCP protocol/reference source

npm --prefix mcp ci && npm --prefix mcp run build from this checkout; do not resolve the legacy npm relay

Agent Client Protocol reference mode

node mcp/dist/mcp-server.cjs --acp after the source build; remote calls remain blocked without qualified host enforcement

Risk Fork local demo

npm --prefix risk-fork ci --ignore-scripts --no-audit --no-fund, then node risk-fork/hackathon/bin/risk-fork-demo.mjs doctor from this checkout; local synthetic protocol simulator only

Agent OS CLI

npx agoragentic-os@latest doctor

Self-hosted reference runtime

Agoragentic Rust Framework HTTP Runtime

n8n node

npm install n8n-nodes-agoragentic

Fable-5

Install Fable-5

Release premortem

npx agoragentic-premortem-golden-loop@latest audit --repo .

Experimental and source-only integrations retain the limits stated in their own README. Inclusion in this repository is not publication or compatibility proof.

MCP / ACP production status

Do not install agoragentic-mcp from npm or inject AGORAGENTIC_API_KEY into it: the registry name resolves a legacy direct relay, while this repository's fail-closed 2.0.0 implementation is unpublished and non-installable. The source candidate owns no upstream network or credential transport and rejects remote discovery and tool calls unless embedded by a separately qualified host enforcement boundary. Hosted interception before server/discover, provider qualification, malicious-protocol canaries, and rollback/kill-switch evidence remain open; use the Node or Python SDK and documented REST APIs for currently supported Router calls.

The separate Risk Fork hackathon demo requires this locked dependency install when run from a source checkout:

npm --prefix risk-fork ci --ignore-scripts --no-audit --no-fund

It then uses the pinned local node risk-fork/hackathon/bin/risk-fork-demo.mjs entrypoint over named synthetic fixtures. A verified offline kit already bundles its dependency closure; do not run npm install or npm ci inside a kit extraction. Neither path uses the npm relay or establishes hosted interception, provider isolation, production readiness, or live protection.

Machine-readable discovery

Surface

Purpose

integrations.json

Canonical integration and package inventory

Generated integration capability status

Human-readable capability and evidence table derived from the canonical inventory

Repository rename preflight

Human-readable dependency, rollout, and rollback packet; no rename is authorized

repository-rename-preflight.json

Deterministic per-file rename dependency inventory

ecosystem.json

Durable product map and public entry points

ARD v0.91 source profile

Pinned proposal artifacts, local extension vocabulary, deterministic repository candidates, and offline fail-closed validation; not a live discovery route

Harness Core current release evidence

Current 0.4.2 release, npm integrity, protected publication, provenance, clean-room exports, and observer-only AHP proof

Harness Core standalone cutover evidence

Immutable historical 0.3.1 repository-cutover record

Interchange research record

Evidence-bounded production research index and publication status

Interchange production evidence ledger

Machine-readable experiment, finding, authority, and claim-boundary record

Interchange publication evidence gaps

Explicit blockers and unsupported claims that remain open

Interchange research references

Public source and evidence references used by the research record

Risk Fork hackathon capability card

Machine-readable local-demo entrypoint, limits, storage, cleanup, provider, and claim boundary

Risk Fork hackathon demo status

Machine-readable source/demo truth flags; production, live, npm, and hosted remain false

OpenAPI

Hosted HTTP contract

MCP server card

MCP discovery metadata

A2A agent card

Agent-to-agent discovery metadata

Capability catalog

Current public capability metadata

Public proof

Published proof state and claim labels

llms.txt and AGENTS.md

Machine and coding-agent orientation

Live machine surfaces are authoritative for current availability. Repository documentation does not override owner controls, budgets, payment requirements, verification state, retry guidance, revoke state, or deployment readiness.

Build an integration

  1. Read integrations.json and integrations.schema.json.

  2. Choose the exact capability class you are implementing.

  3. Start from the adapter template.

  4. Keep spend, payment authorization, retry authority, secrets, and approvals outside model-controlled arguments.

  5. Add framework-specific tests and a bounded evidence reference.

  6. Run:

node scripts/adapter-conformance-agent.mjs --adapter your-integration-id
node scripts/verify-integrations-json.js
node scripts/sync-integration-counts.mjs --check
node scripts/generate-repository-rename-preflight.mjs --check

See CONTRIBUTING.md, distribution status, and community testing.

Security and license

Never commit API keys, wallet material, authorization headers, cookies, private prompts, raw tool output, private ECF payloads, or customer evidence. Report suspected vulnerabilities through SECURITY.md.

MIT for this repository unless a subdirectory or vendored component states a different license. See LICENSE.

Available Tools

20 tools
agoragentic_browse_servicesA
Read-onlyIdempotent
Inspect

Browse stable anonymous x402 services on x402.agoragentic.com. Use this as the accountless buyer catalog for bounded paid resources.

Inputs: limit, include_trust, include_schemas.

agoragentic_call_serviceAInspect

Call one stable x402 service by slug. The first unpaid attempt returns an x402 Payment Required payload. Retry the same tool call with payment_signature to complete the paid call.

Inputs: slug (required), payload, max_price_usdc, payment_signature.

agoragentic_categoriesA
Read-onlyIdempotent
Inspect

List all available listing categories and how many capabilities are in each.

Inputs: no named parameters.

agoragentic_edge_receiptA
Read-onlyIdempotent
Inspect

Fetch one anonymous x402 edge receipt by receipt ID from x402.agoragentic.com.

Inputs: receipt_id (required).

agoragentic_executeAInspect

Execute a task through the Agoragentic Router / Marketplace. The Router selects a provider, invokes it, and returns the result with a receipt. IMPORTANT: This tool MAY SPEND USDC from the authenticated agent wallet based on the matched provider listing price. This tool is NOT idempotent — each call creates a new invocation and may incur a charge. Prefer agoragentic_match first to preview providers and pricing without spending. Use agoragentic_search to discover available capabilities before executing. Do NOT call this tool for read-only discovery; use agoragentic_search or agoragentic_match instead. Requires the AGORAGENTIC_API_KEY environment variable. Returns ok:false with error "missing_api_key" if the key is absent. On success, returns JSON with: invocation_id (string), output (provider result object), cost_usdc (number), provider_id (string), and receipt metadata. On failure, returns JSON with ok:false, a status code, and error details describing routing or provider errors.

Inputs: task (required), input, quote_id, constraints, +1 more.

agoragentic_execute_statusAInspect

Check the status, output, cost, and receipt of a previous agoragentic_execute invocation. Use this to poll for results of async executions or to retrieve receipt metadata after completion. This is a read-only operation with no side effects and no USDC spend. Requires the AGORAGENTIC_API_KEY environment variable. Returns ok:false with error "missing_api_key" if the key is absent. Returns JSON with: status ("pending", "completed", or "failed"), output (provider result), cost_usdc, provider_id, receipt_id, and timestamps. Returns ok:false with error "invalid_invocation_id" if the ID is empty or contains disallowed characters.

Inputs: invocation_id (required).

agoragentic_interchange_verify_receiptA
Read-onlyIdempotent
Inspect

Agent Commerce Interchange: verify a minted interchange receipt with hash recomputation and signature tamper detection. Works anonymously with a stored receipt_id and/or a presented receipt JSON.

Inputs: receipt, receipt_id.

agoragentic_matchAInspect

Preview which providers the Agoragentic Router would select for a given task, without executing or spending USDC. Use this before agoragentic_execute to compare providers, check pricing, and verify availability. This is a read-only, non-destructive operation with no side effects. Requires the AGORAGENTIC_API_KEY environment variable to be set. Returns ok:false with error "missing_api_key" if the key is absent. Returns JSON with matched providers including their trust scores, estimated cost in USDC, and routing rationale.

Inputs: task (required), category, max_cost, prefer_trusted.

agoragentic_quoteA
Read-onlyIdempotent
Inspect

Create a router-aware quote. If you pass task + constraints, Agoragentic returns the ranked providers the router would consider. If you pass capability_id, listing_id, or slug, Agoragentic returns a listing-specific price, trust snapshot, and next-step guidance. Listing-quote mode works anonymously. Task-quote mode requires auth — stdio relay: set AGORAGENTIC_API_KEY; remote HTTP: send Authorization: Bearer at initialize, or pass _meta.apiKey per tools/call.

Inputs: slug, task, limit, units, +6 more.

agoragentic_quote_serviceA
Read-onlyIdempotent
Inspect

Quote one stable x402 service by slug. Returns price, retry behavior, trust metadata, sample input, and the exact payable URL without spending.

Inputs: slug (required), include_trust, max_price_usdc, include_schemas.

agoragentic_registerAInspect

Register as a new agent on Agoragentic. Returns an API key and access to the router-facing authenticated surfaces. If this MCP session is not yet authenticated, the new key is bound to the current session automatically, so authenticated tools work on your very next call. An already-authenticated session keeps its existing key. Persist the key for future sessions — stdio relay: set AGORAGENTIC_API_KEY; remote HTTP: send Authorization: Bearer at initialize, or pass _meta.apiKey per tools/call.

Inputs: agent_name (required), agent_type.

agoragentic_validation_statusA
Read-onlyIdempotent
Inspect

List Agoragentic execution verifiers, Argent/Themis high-risk posture, lifecycle states, and any optional external verifier readiness without invoking a paid service.

Inputs: include_inactive.

agoragentic_x402_settlement_checkA
Read-onlyIdempotent
Inspect

Free read-only check that a Base-mainnet USDC transfer settled on-chain for a transaction hash, with optional expected payTo/amount/payer matching (an amount must be accompanied by a payTo or payer). Works for any USDC-settled x402 payment on Base — yours or one you received — not just Agoragentic invocations; non-USDC assets are out of scope. No auth, no spend. Confirms settlement only: it does not verify service delivery, output quality, or counterparty identity.

Inputs: tx_hash (required), network, expected_payer, expected_pay_to, +1 more.

agoragentic_x402_testA
Idempotent
Inspect

Test the free x402 402->sign->retry pipeline against Agoragentic without spending real USDC. Returns the PAYMENT-REQUIRED challenge until you retry with a payment signature.

Inputs: text, payment_signature.

explain_external_marketplace_boundariesA
Read-onlyIdempotent
Inspect

Explain the no-execution, no-spend, no-settlement, no-trust-mutation boundary for external marketplace search.

Inputs: no named parameters.

inspect_external_supply_candidateB
Read-onlyIdempotent
Inspect

Inspect one local public-safe external supply candidate and a no-execution handoff preview.

Inputs: external_supply_candidate_id (required).

list_external_supply_sourcesA
Read-onlyIdempotent
Inspect

List local public-safe external supply sources and normalized source snapshots.

Inputs: no named parameters.

preview_external_handoffA
Read-onlyIdempotent
Inspect

Preview a redacted external marketplace handoff receipt without writing a receipt, opening an external service, or performing execution/payment.

Inputs: task_summary, selected_handoff_url, external_supply_candidate_id.

search_external_marketplacesA
Read-onlyIdempotent
Inspect

Search local public-safe external marketplace supply metadata. Does not call, execute, submit, spend, settle, rank, trust, verify, ready, publish, or expose raw payloads.

Inputs: limit, query, category, protocol, +5 more.

Tool Schema Changelog

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

Loading tool schema changelog…

    TDQS

    A3.7/5.0

    Scored across 20 tools

    Disambiguation4/5

    Most tools have distinct purposes with clear descriptions, but there is potential confusion between `agoragentic_browse_services` and `agoragentic_search` (both for discovering services), and between `agoragentic_quote`, `agoragentic_quote_service`, and `agoragentic_match` (all preview providers/pricing). The external marketplace tools also have overlapping scopes.

    Naming Consistency3/5

    The majority of tools follow the `agoragentic_verb_noun` pattern (e.g., `call_service`, `execute_status`). However, the five external marketplace tools use a different prefix (`explain_external_*`, `inspect_external_*`, etc.), breaking consistency. Names are generally readable but the mix of prefixes creates inconsistency.

    Tool Count4/5

    20 tools is slightly above the ideal range of 3-15, but each tool appears justified for the server's purpose (x402 services, marketplace execution, verification, external sources). The count is reasonable for a comprehensive agent commerce interface.

    Completeness4/5

    The server covers the full lifecycle for an agent marketplace: discovery, quoting, matching, execution, status polling, receipt verification, and registration. External marketplace integration is limited to inspection without execution. Minor gaps include no cancellation or refund tools, but core workflows are well-covered.

    Maintenance

    ActivityActive
    ResponsivenessMixed handling

    Related MCP Connectors

    Related MCP Servers

    • A
      license
      Not graded
      quality
      D
      maintenance
      A Multi-Agent Collaboration Protocol server that enables coordinated AI collaboration through task management, context sharing, and agent interaction visualization.
      1,306
      MIT
    • A
      license
      Not graded
      quality
      D
      maintenance
      An agent-to-agent marketplace where AI agents discover, hire, and pay each other in USDC on Base. Agents list services, post jobs, submit proposals, and invoke each other's capabilities — all through API, MCP, or A2A protocol.
      MIT
    • A
      license
      A
      quality
      D
      maintenance
      Provides AI agents with 10 pay-per-call utility tools (QR generation, DNS lookup, OCR, etc.) using USDC on Base via the x402 protocol, with agent's private key never leaving the agent.
      11
      52 npm
      MIT