Protocol positioning
Where ASP fits in the protocol landscape alongside MCP and A2A
ASP occupies a specific position in the agent protocol stack: the session and presentation layers (OSI Layers 5-6) between application protocols like MCP and A2A above, and transport protocols like npayload below. This page explains that positioning in detail, compares ASP to both modern and historical protocols, and shows how they work together.
OSI layer positioning
ASP is designed to fill the session and presentation layers of the OSI model. This is a deliberate architectural choice. Layer 7 protocols define what agents do. Layer 1-4 protocols define how messages move. ASP defines how agents conduct structured conversations in between.
| OSI Layer | Function | Protocol | What it provides |
|---|---|---|---|
| Layer 7 | Application | MCP | Tool discovery and invocation for language models |
| Layer 7 | Application | A2A | Task delegation, lifecycle management, and agent discovery |
| Layer 6 | Presentation | ASP | Trust scoring, context serialization, hash chain integrity, message signing |
| Layer 5 | Session | ASP | Session lifecycle, turn-taking, negotiation, performatives, commitments |
| Layer 4 | Transport | npayload | Reliable delivery, ordering, retries, dead-letter queues |
| Layer 3 | Network | npayload | Cross-region routing, instance fabric, interest-based routing |
| Layer 2 | Data Link | npayload | Channel-level framing, subscription management |
| Layer 1 | Physical | Infrastructure | Network, compute, storage |
Understanding the layer split
ASP spans two OSI layers because agent sessions require both session management and data representation:
- Layer 5 (Session) handles the conversation structure: opening and closing sessions, managing phases, enforcing turn-taking, tracking which performative is valid at which point in the lifecycle.
- Layer 6 (Presentation) handles data representation: trust score computation, context object serialization, SHA-256 hash chain construction, and Ed25519 signature verification.
These two concerns are tightly coupled in practice. A trust score (Layer 6) determines which session transitions (Layer 5) are permitted. A hash chain (Layer 6) ensures message integrity across session phases (Layer 5). Splitting them into separate protocols would create unnecessary complexity.
Protocol comparison
The following table compares ASP with both modern agent protocols and historical predecessors across seven dimensions.
| Dimension | ASP | MCP | A2A | FIPA ACL | SIP |
|---|---|---|---|---|---|
| OSI Layer | 5-6 | 7 | 7 | 5-7 | 5 |
| Purpose | Session management, negotiation, trust, commitments | Tool discovery and invocation | Task delegation and lifecycle | General agent communication | Real-time session setup |
| Session support | Eight-phase lifecycle with explicit state transitions and configurable timeouts | Stateless request-response. No session concept. | Partial. Task lifecycle (submitted, working, completed, failed) but no negotiation phases. | Complex conversation protocols with multiple interaction patterns (FIPA-Request, FIPA-Contract-Net) | Dialog lifecycle with INVITE, ACK, BYE, and session timers |
| Trust model | Eight-component quantitative model with time decay, growth curves, and peer evaluation | None. Relies on server-side configuration to control tool access. | None. Static Agent Cards advertise capabilities but do not quantify reliability. | None. Assumes a trusted communication channel exists. | None. Relies on TLS for transport-level security. |
| Commitment model | Binding commitments with structured terms, deadlines, escrow, penalties, and dispute resolution | None | None | None | None |
| Multi-party | N-party sessions with role-based participation (initiator, responder, observer) | Client-server only (one client, one server) | Supports multi-agent task delegation chains | Supports multi-party interaction protocols | Supports multi-party conferencing via SDP |
| Transport | Transport-agnostic. Primary binding to npayload. HTTP/SSE and WebSocket bindings available. | JSON-RPC over stdio, HTTP/SSE, or WebSocket | HTTP with JSON payloads, Server-Sent Events for streaming | Originally CORBA/IIOP. Later HTTP bindings added. | UDP, TCP, or TLS |
ASP is the only protocol in this comparison that provides all three of: quantitative trust, binding commitments, and structured negotiation. The others address subsets of these concerns, or none at all.
When to use which
The protocols are complementary, not competing. Each answers a different question about agent interaction.
| You need to... | Use | Why |
|---|---|---|
| Discover and invoke tools on a remote system | MCP | MCP defines tools/list and tools/call for structured tool access |
| Delegate a task to another agent | A2A | A2A provides task lifecycle with status tracking and streaming |
| Negotiate terms before acting | ASP | ASP's PROPOSE, COUNTER, ACCEPT, REJECT performatives formalize negotiation |
| Verify whether an agent is trustworthy | ASP | ASP's eight-component trust model quantifies reliability from observed behaviour |
| Create a binding commitment with penalties | ASP | ASP commitments include terms, deadlines, escrow amounts, and dispute resolution |
| Coordinate a conversation among three or more agents | ASP | ASP sessions support N-party participation with role-based access |
| Deliver messages reliably across regions | npayload | npayload handles retries, DLQ, encryption, and cross-instance routing |
| Escalate to a human when agents cannot resolve an issue | ASP | The ESCALATE performative transfers full session context to a human operator |
Understanding protocol boundaries
A common question is where one protocol ends and another begins. The boundaries are clean:
- MCP operates within a single request-response cycle. "Call this tool with these parameters, return the result." There is no concept of a preceding negotiation or a subsequent trust update.
- A2A operates within a task lifecycle. "Here is a task. Work on it. Tell me when it is done." There is no concept of terms, penalties, or dispute resolution.
- ASP operates across the full conversation. "Let us establish a session, negotiate terms, form commitments, execute, verify, learn, and close." MCP calls and A2A tasks happen inside this container.
Complementary usage: procurement example
In practice, all three protocols work together. Here is a detailed example of a procurement negotiation between a purchasing agent and three supplier agents.
ASP: Open session and discover participants
The purchasing agent opens an ASP session and invites three supplier agents. Each supplier joins as a responder. Session parameters are set: 30-minute timeout, maximum 20 turns per participant, minimum trust threshold of 0.6 for commitment eligibility.
{
"performative": "INFORM",
"phase": "DISCOVERY",
"context": {
"sessionType": "competitive-bid",
"itemCategory": "industrial-sensors",
"quantity": 5000,
"deliveryRegion": "EU-West"
}
}MCP: Discover supplier capabilities
The purchasing agent uses MCP to call tools/list on each supplier. Supplier A exposes check_inventory, get_bulk_pricing, and create_order. Supplier B exposes get_pricing and create_order but not check_inventory. Supplier C exposes all three plus get_certifications. These capabilities inform the negotiation strategy.
ASP: Trust assessment
Before negotiating, the purchasing agent checks each supplier's ASP trust scores. Supplier A has a reliability score of 0.92 and latency score of 0.87. Supplier B scores 0.78 on reliability. Supplier C scores 0.95 on reliability but 0.61 on latency. The purchasing agent weights reliability at 2x and proceeds with all three.
{
"performative": "OBSERVE",
"phase": "TRUST_ASSESSMENT",
"trustScores": {
"supplierA": { "composite": 0.89, "reliability": 0.92, "latency": 0.87 },
"supplierB": { "composite": 0.76, "reliability": 0.78, "latency": 0.81 },
"supplierC": { "composite": 0.82, "reliability": 0.95, "latency": 0.61 }
}
}ASP: Negotiate with PROPOSE and COUNTER
The purchasing agent sends a PROPOSE to all three suppliers: 5,000 units at $8.00 each, delivery within 10 business days, 1.5% daily penalty for late delivery, 500-token escrow.
Supplier A sends a COUNTER: $9.25 per unit, 7-day delivery, 1% penalty. Supplier C sends a COUNTER: $8.50, 14-day delivery, 0.5% penalty. Supplier B ACCEPTs the original terms.
The purchasing agent evaluates and ACCEPTs Supplier A's counter-offer based on the combination of price, speed, and trust score.
ASP + A2A: Commit and delegate execution
The purchasing agent and Supplier A exchange COMMIT performatives, creating a binding commitment with the agreed terms. The purchasing agent then delegates order execution to Supplier A via A2A, referencing the ASP session ID and commitment ID.
{
"performative": "COMMIT",
"phase": "COMMITMENT",
"terms": {
"quantity": 5000,
"unitPrice": 9.25,
"deliveryDays": 7,
"penaltyRate": 0.01,
"escrow": 500
},
"commitmentId": "asc_k7m9x2p4"
}ASP: Verify, learn, and close
Supplier A completes the order. The purchasing agent verifies fulfilment against the commitment terms. Trust scores are updated: Supplier A's reliability increases to 0.93. The session enters the LEARN phase where the purchasing agent records that Supplier A's counter-offers are typically 15% above ask but deliver faster. The session closes.
Comparison with historical protocols
ASP draws from four traditions in protocol design. Understanding these predecessors clarifies why ASP makes the choices it does.
Understanding FIPA ACL
The Foundation for Intelligent Physical Agents published an Agent Communication Language (FIPA ACL, IEEE 2002) based on speech act theory. FIPA's performatives (INFORM, REQUEST, PROPOSE, ACCEPT-PROPOSAL) directly influenced ASP's thirteen performatives.
ASP differs from FIPA ACL in three significant ways:
| Dimension | FIPA ACL | ASP |
|---|---|---|
| Trust | Assumed a trusted channel. No quantitative trust model. | Eight-component trust scoring computed from observed behaviour. |
| Economics | No economic mechanisms. No penalties, escrow, or budget tracking. | Binding commitments with escrow, penalties, and dispute resolution. |
| Transport | Designed for CORBA/IIOP. Later HTTP bindings were added as afterthoughts. | Transport-agnostic from day one, with first-class bindings for npayload, HTTP/SSE, and WebSocket. |
Understanding SIP (RFC 3261)
SIP pioneered session lifecycle management for real-time communication. Its dialog model (INVITE, ACK, BYE, session timers) directly influenced ASP's eight-phase lifecycle. Both protocols treat sessions as first-class objects with explicit setup, maintenance, and teardown.
ASP extends SIP's model with negotiation semantics, trust scoring, and commitment tracking that real-time voice and video calls do not require. SIP sessions last minutes. ASP sessions may last hours, days, or weeks.
Understanding OAuth 2.0 and DPoP (RFC 9449)
OAuth 2.0 and DPoP established patterns for identity binding and cryptographic proof-of-possession. ASP applies these patterns to agent identity within sessions: session participants are cryptographically bound to their messages via Ed25519 signatures, preventing impersonation and message replay.
The key insight from DPoP is that possession of a token is insufficient proof of identity. The proof must be bound to the specific request. ASP applies this principle at the session level: each message is signed, and each signature is bound to the session context and hash chain.
Understanding economic game theory
ASP's commitment model draws from mechanism design theory: the study of designing rules so that rational agents behave in desirable ways. Binding commitments with penalties, escrow mechanisms, and dispute resolution create credible consequences for defection, making cooperation the rational strategy.
Without economic mechanisms, agent cooperation relies on good faith. With them, cooperation becomes the equilibrium outcome regardless of whether agents are "well-behaved" by design.
Conformance levels
Not every implementation needs every feature. ASP defines three conformance levels:
| Level | Requirements | Use case |
|---|---|---|
| Basic | INFORM, QUERY, CLOSE performatives. Session open/close. No trust scoring. | Simple information exchange between known agents |
| Standard | All 13 performatives. Eight-phase lifecycle. Trust scoring. No commitments. | Multi-turn conversations with trust verification |
| Full | Standard plus binding commitments, escrow, dispute resolution, and LEARN phase. | High-value negotiations requiring economic accountability |
An agent at the Basic conformance level can participate in a session with a Full-conformance agent. The Basic agent simply will not use performatives it does not support. ASP is designed so that capability differences do not prevent interoperability.
Next steps
Was this page helpful?