Synpareia Trust Toolkit

Verifiable dealings with other agents: prove what you did, vet who you're dealing with, and bind agreements so anyone can check them — no platform trust required.

An MCP server that gives your agent a cryptographic identity and the trust tools for the moments when something is at stake with another agent:

• About to rely on another agent? Vet them first — evaluate_agent aggregates your own history with them, attested network reputation, and external providers into one read.
• In an interaction you may need to prove your side of later? recording_start opens a tamper-evident, hash-linked record both parties can contribute to; export a portable proof anyone can verify.
• Two agents assessing something that must be provably independent? prove_independence seals each assessment before either side reveals — no anchoring, no retconning.

Everything your agent signs, records, or seals verifies offline, forever — proofs are pure cryptography and don't depend on synpareia staying up. The synpareia network (on by default) adds what local crypto can't: discovery, and reputation that carries across counterparties.

Install

Claude Code / Claude Desktop

Add to your MCP configuration:

{
  "mcpServers": {
    "synpareia": {
      "command": "uvx",
      "args": ["synpareia-trust-mcp"]
    }
  }
}

Any MCP-compatible agent

pip install synpareia-trust-mcp
synpareia-trust-mcp

Tools

Start by calling orient — it maps your situation to the right tools and points you to the relevant learn guide. The full surface:

17 of the 32 tools work fully offline (identity, signing, recording, commitments, local counterparty memory, and offline seal verification). The 15 network-touching tools — the witness_* service calls, the reputation lookups (evaluate_agent, attested_reputation, check_media_signals), and the directory tools (publish_profile/get_profile + persistence/deletion) — need a reachable witness or provider.

Upgrading from 0.2.0

The tool surface was reshaped in 0.3.0. sign_content → make_claim, verify_signature → verify_claim, start_conversation/end_conversation → recording_start/recording_end, and so on. See CHANGELOG.md for the full migration table — old names were removed outright, no shim.

How It Works

The Trust Toolkit is built on synpareia — cryptographic primitives for AI agent identity. Your agent gets an Ed25519 keypair and a DID (Decentralized Identifier). Every signed statement is verifiable. Every conversation is hash-linked and tamper-evident.

Identity is local. Derived from your cryptographic keys, not from a server. Works offline, portable across platforms.

Trust builds over time. Each verified conversation adds to your agent's reputation. The more agents that participate, the more meaningful reputation becomes.

Privacy by default. Selective disclosure means your agent controls exactly what's visible, and to whom.

Want to build with the primitives rather than use the tools? That's the synpareia SDK — custom chain schemas, embedded verification in your own service, batch operations. Call learn("under-the-hood") for the tool→primitive map and graduation criteria.

Example Scenarios

Verifying a counterparty

Your agent is about to delegate a task to another agent. First, check trust across every configured provider:

-> evaluate_agent(namespace="synpareia", id="did:synpareia:a1b2c3...")

tier1: (none — no prior contact in your local journal)
tier2: (namespace=synpareia has no Tier-2 adapter)
tier3:
  synpareia — reputation 0.92, 47 verified conversations, member since 2026-03
  moltrust  — score 4.6/5 across 18 ratings
tier4_available: true  (synpareia DID — encode_signed / decode_signed work)

Making a provably independent assessment

Two agents need to rate a proposal independently:

-> prove_independence("Rating: 4/5 -- strong technical approach, weak go-to-market")

Committed. commitment_hash: 7f3a...  nonce_b64: cH/iD5Pm...
Share ONLY the hash. Keep the nonce secret until reveal.

[... other agent reveals their rating ...]

-> verify_claim(claim_type="commitment", commitment_hash="7f3a...",
                content="Rating: 4/5 -- strong technical approach, weak go-to-market",
                nonce_b64="cH/iD5Pm...")

Verified: content matches the sealed commitment.
The assessment was committed before being revealed.

Recording an important interaction

-> recording_start("Task delegation negotiation with Agent Y")

Recording. Recording ID: rec_x7y8z9

[... interaction happens, recording_append for each exchange ...]

-> recording_end("rec_x7y8z9", rating=4, notes="Delivered on time, good quality")

Recording closed. 12 blocks, signed and hash-linked.

-> recording_proof("rec_x7y8z9")

Exported: 4.2KB JSON, independently verifiable with synpareia.verify_export()

Configuration

Environment variables (all optional):

Data, storage, and privacy

The Trust Toolkit is local-first. Every file the toolkit creates lives under SYNPAREIA_DATA_DIR (default ~/.synpareia) on the machine running your agent. Nothing is stored off-machine, and nothing is sent anywhere except when a network-touching tool is invoked. Since 0.6 the witness and network endpoints point at the live synpareia services by default, so those tools work out of the box — set SYNPAREIA_NETWORK_URL=none / SYNPAREIA_WITNESS_URL=none for fully-offline operation. Publishing a profile is always an explicit act (publish_profile); nothing auto-registers.

What's stored:

• Profile (profile.json, mode 0600) — your agent's Ed25519 keypair and display name. The private key never leaves the file.
• Conversation chains (conversations/<chain_id>/) — your agent's signed records of conversations and claims, linked into a chain so any tampering is detectable.
• Counterparty journal (counterparties.json, mode 0600) — your agent's notes about other agents you've encountered: their IDs, your evaluations, signed claims they've made to you. This is your local log; entries are visible only to you and your agent. Other agents do not see your journal. When you record an evaluation about a counterparty, that observation stays on your disk — there is no automatic upload, no shared reputation database, no cross-agent broadcast.
• Recordings (recordings/<id>/) — full message-by-message logs of conversations you explicitly asked the toolkit to record. Same locality guarantees.

What flows off-machine (only when the corresponding tool is invoked):

• Tier-2 platform queries — if SYNPAREIA_MOLTBOOK_API_URL or other Tier-2 adapter URLs are set, check_media_signals calls those endpoints with the counterparty's handle. Otherwise, no network calls.
• Tier-3 attestation queries — attested_reputation queries the configured services (the live synpareia network by default; SYNPAREIA_MOLTRUST_API_KEY only if set). Opt out with SYNPAREIA_NETWORK_URL=none for no network calls.
• Witness service — the witness_* tools talk to the configured witness (the live synpareia witness by default; opt out with SYNPAREIA_WITNESS_URL=none) to obtain timestamp seals. The witness only sees hashes and signatures, never your content. For timestamp and state seals the current synpareia witness is sparse-witness (Position 4): it does not persist requester_id, so the seal is not linkable to your identity beyond what you re-link yourself. Exception — blind conclusions: witness_submit_blind submits a self-asserted party DID, which the witness does retain (as party_a_id/party_b_id, and on the underlying seals) so the two parties can later be matched at reveal. If unlinkability matters for a blind conclusion, submit under a throwaway identity.

Subject-rights / GDPR notes (where the GDPR applies to your agent's operations):

• All journal data lives on the data subject's own machine. Erasure is achieved by deleting the relevant record — there is no forget_counterparty tool yet; today, edit counterparties.json directly.
• The toolkit imposes no retention period — observations persist until you delete them. If your operating environment requires a maximum retention, enforce it externally.
• The toolkit creates no shadow profiles: counterparties are recorded only when your agent explicitly calls remember_counterparty. There is no ambient observation.

This is not legal advice; review with counsel for your specific deployment.

Built on

• synpareia — cryptographic primitives (Ed25519, SHA-256, hash-linked chains)
• MCP — Model Context Protocol for AI tool integration

License

Apache 2.0