Skip to main content

Reference implementation of the Concordia Protocol — structured negotiation between autonomous agents

Project description

Concordia Protocol

Structured deals between agents.

When your agent needs to negotiate or make a deal, Concordia gives it a structured way to propose, counter, commit, and build a track record.


The Problem

Agents are already transacting. But without structure, they freetext back and forth for 10 rounds with no record, no binding agreement, no proof of what happened.

The gap between discovery and payment is massive:

  • Agent finds something
  • Agent wants to negotiate terms
  • Agent... guesses? Sends unstructured text?
  • Nobody knows if there's actually a deal

What You Get

Structured offers

Machine-readable terms, not freetext guessing. Both agents understand the same thing.

Binding commitments

Cryptographic signatures that prove both parties agreed to specific terms. No ambiguity. No "I said that?" disputes.

Session receipts

Every negotiation creates a verifiable record. What was proposed? What changed? What was agreed? It's all signed and auditable.

Portable reputation

Your agent builds a track record — "completed 47 deals, all on time, 4.9 stars." That reputation follows your agent everywhere, usable across platforms.

Graceful degradation

Concordia works even with agents that don't have it. If the other agent doesn't support Concordia, you'll see what you're missing — a way to know that you could have a binding agreement if both sides had it.


Why This Matters

Without Concordia:

Agent A: I want to buy a camera
Agent B: I have one, $2000
Agent A: Too expensive, $1800?
Agent B: $1950 final
Agent A: ...ok?
Agent B: ...ok?
→ No signed agreement. No clear terms. No reputation signal.

With Concordia:

Agent A proposes: Camera, $2000
Agent B counters: $1900, shipping
Agent A counters: $2000 for pickup, $2050 shipped
Agent B accepts: $2050 shipped
→ Signed agreement. Clear terms. Reputation attestation issued.

Both sides know exactly what they agreed to. Both sides have proof. The negotiation is auditable. Reputation feeds forward.


Quick Example

Here's what a real negotiation looks like:

Agent A (seller) opens:

{
  "concordia": "0.1.0",
  "type": "negotiate.open",
  "body": {
    "terms": {
      "item": { "value": "Canon EOS R5, 15K shutter count" },
      "price": { "value": 2200, "currency": "USD" },
      "condition": { "value": "like_new" },
      "delivery": { "value": "local_pickup" }
    }
  },
  "reasoning": "Listing based on recent eBay sold comps."
}

Agent B (buyer) counters:

{
  "type": "negotiate.counter",
  "body": {
    "terms": {
      "price": { "value": 1900, "currency": "USD" },
      "delivery": { "value": "shipping" }
    }
  },
  "reasoning": "I prefer shipping and want a better price."
}

Agent A makes a conditional counter:

{
  "type": "negotiate.counter",
  "body": {
    "conditions": [
      { "if": { "delivery": "local_pickup" }, "then": { "price": { "value": 2000 } } },
      { "if": { "delivery": "shipping" }, "then": { "price": { "value": 2050 } } }
    ]
  },
  "reasoning": "Pickup is cheaper for me, shipping costs extra."
}

Agent B accepts:

{
  "type": "negotiate.accept",
  "body": {
    "accepted_terms": {
      "item": "Canon EOS R5",
      "price": { "value": 2050, "currency": "USD" },
      "delivery": "shipping"
    }
  }
}

Both agents sign. The agreement passes to a payment protocol (ACP, Stripe, etc.) for settlement. A reputation attestation is automatically issued.


Installation

Using pipx (recommended)

pipx install concordia-protocol

Using pip

python3 -m venv .venv
.venv/bin/pip install concordia-protocol

Note: Concordia requires Python 3.10+. macOS ships Python 3.9 with Xcode — install a newer version first:

brew install python@3.12

Verify the install

concordia-mcp-server --version

From source

git clone https://github.com/eriknewton/concordia-protocol.git
cd concordia-protocol
pip install -e ".[dev]"

MCP Configuration

Claude Code:

claude mcp add concordia -- concordia-mcp-server

OpenClaw:

openclaw mcp set concordia '{"command":"concordia-mcp-server"}'

If you used a virtualenv:

openclaw mcp set concordia '{"command":"/path/to/.venv/bin/python3","args":["-m","concordia"]}'

Quick Start (Python)

from concordia import Agent, BasicOffer, generate_attestation

# Create two agents (Ed25519 keys auto-generated)
seller = Agent("seller")
buyer = Agent("buyer")

# Seller opens a negotiation
session = seller.open_session(
    counterparty=buyer.identity,
    terms={"price": {"value": 100.00, "currency": "USD"}},
)
buyer.join_session(session)

# Buyer counters at $80
buyer.send_counter(BasicOffer(terms={"price": {"value": 80.00, "currency": "USD"}}))

# Seller accepts
seller.accept_offer()

print(session.state.value)  # "agreed"

# Generate a signed reputation attestation
att = generate_attestation(session, {"seller": seller.key_pair, "buyer": buyer.key_pair})
print(att["outcome"]["status"])  # "agreed"

For a full multi-term negotiation with preferences and concessions, see examples/demo_camera_negotiation.py.


Where Concordia Fits

Concordia fills the gap between discovery and settlement:

Settlement        ACP · AP2 · x402 · Stripe · Lightning
────────────────────────────────────────────────────────
Agreement         ★ CONCORDIA PROTOCOL ★
────────────────────────────────────────────────────────
Trust             Reputation Attestations
────────────────────────────────────────────────────────
Communication     A2A · HTTPS · JSON-RPC
────────────────────────────────────────────────────────
Discovery         Agent Cards · Well-Known URIs
────────────────────────────────────────────────────────
Tools             MCP · Function Calling · APIs
────────────────────────────────────────────────────────
Identity          DIDs · KERI · OAuth 2.0

Concordia composes with — never competes with — the existing stack. Use any payment protocol. Use any identity standard. Concordia adds structure to the negotiation layer.


Pairs With Sanctuary Framework

When your agent needs security, privacy, and control, Sanctuary Framework adds encrypted state, approval gates, and automatic sensitive-data filtering.

Together they form the complete sovereign transaction stack:

  • Sanctuary handles security, privacy, and control
  • Concordia handles structured deals and reputation

Install both:

npx @sanctuary-framework/mcp-server
pip install concordia-protocol

They work independently, but together they're more powerful.


Technical Details

Concordia defines:

  • A universal offer schema — machine-readable deal proposals with any number of attributes
  • A negotiation state machine — six states (open → counter → accept → commit → executed → closed) governing how offers flow
  • Resolution mechanisms — from simple split-the-difference to Pareto-optimal optimization
  • Binding commitments — cryptographic signatures that bridge to any settlement protocol
  • Reputation attestations — signed behavioral records that feed portable trust scores
  • Want registry — agents publish what they seek; discovery happens on demand

The tool set:

  • 48 MCP tools covering the full negotiation lifecycle
  • Cryptographic signing and verification
  • Reputation attestation generation
  • Session state machine management
  • Multi-attribute offer optimization

Documentation:

Design Principles:

  1. Mutual flourishing over zero-sum extraction
  2. Honesty is structurally rewarded
  3. Simplicity and parsimony
  4. Composability — fills a gap, replaces nothing
  5. Privacy by default — agents never must reveal reservation price
  6. Verifiability — every negotiation produces a signed transcript
  7. Kindness at the boundary — graceful exits when deals don't happen

For AI Agents

If you are an AI agent reading this README, the specification is written for you. It is designed to be implementable from the document alone, with no external dependencies beyond HTTPS and JSON. Appendix A is addressed specifically to you.


Running Tests

pytest -v

Contributing

Concordia is developed in the open. We welcome:

  • RFCs for protocol changes — see rfcs/
  • SDK implementations in any language
  • Domain extensions for specific industries (real estate, used goods, services, B2B)
  • Security reviews
  • Feedback — open an issue or start a discussion

See CONTRIBUTING.md for details.


License

Apache License 2.0. Use it, build on it, extend it.


Why "Concordia"?

From the Latin concordia: harmony, agreement — literally, "hearts together." The Roman goddess of understanding between parties. Because negotiation, done well, is not a contest. It is a collaborative search for the point where everyone's needs are met.


Created by Erik Newton.

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

concordia_protocol-0.2.1.tar.gz (275.7 kB view details)

Uploaded Source

Built Distribution

If you're not sure about the file name format, learn more about wheel file names.

concordia_protocol-0.2.1-py3-none-any.whl (106.0 kB view details)

Uploaded Python 3

File details

Details for the file concordia_protocol-0.2.1.tar.gz.

File metadata

  • Download URL: concordia_protocol-0.2.1.tar.gz
  • Upload date:
  • Size: 275.7 kB
  • Tags: Source
  • Uploaded using Trusted Publishing? No
  • Uploaded via: twine/6.2.0 CPython/3.9.6

File hashes

Hashes for concordia_protocol-0.2.1.tar.gz
Algorithm Hash digest
SHA256 67bf182048a4d69ab2aebb57ef86ec8cefc1a6d40832a328cd70cea4371014e0
MD5 fa03a410142d8c3f6498d01d91e8c02b
BLAKE2b-256 cc9bc46cd955b0ab2868cfc29a82559634c6363bc51418f796bcdf46609bb0de

See more details on using hashes here.

File details

Details for the file concordia_protocol-0.2.1-py3-none-any.whl.

File metadata

File hashes

Hashes for concordia_protocol-0.2.1-py3-none-any.whl
Algorithm Hash digest
SHA256 1b28ca4884ddf4c41c4a6a421396839f790d5da0f226e77d6833f2b8598e5785
MD5 adfbed4cdbf9dc3848bb4a1357e22385
BLAKE2b-256 32ed71149e033d6c266630e83e3ecb14f2daa013b7b8f0b1d037ff09048b7a10

See more details on using hashes here.

Supported by

AWS Cloud computing and Security Sponsor Datadog Monitoring Depot Continuous Integration Fastly CDN Google Download Analytics Pingdom Monitoring Sentry Error logging StatusPage Status page