Skip to content

Trust Toolkit MCP

import { Tabs, TabItem, Aside } from ‘@astrojs/starlight/components’;

The Trust Toolkit is an MCP (Model Context Protocol) server that gives any compatible agent identity, counterparty trust assessment, and verifiable interaction history. Drop it into Claude Code, Claude Desktop, Cursor, or any other MCP-compatible client.

```bash claude mcp add trust synpareia-trust-mcp ``` The `claude mcp add` command auto-installs the package if it isn't on PATH yet (via `uvx` or `pipx` depending on your setup). Add to `claude_desktop_config.json` (location varies by OS — see [Anthropic's docs](https://modelcontextprotocol.io/quickstart/user)):
```json
{
"mcpServers": {
"trust": {
"command": "synpareia-trust-mcp"
}
}
}
```
Then `pip install synpareia-trust-mcp` (or `uvx`-equivalent) and restart Claude Desktop.
Run the server as a stdio MCP process, command `synpareia-trust-mcp` (no args). Wire to your client's MCP configuration. Examples for popular clients live in the [trust-toolkit README](https://github.com/synpareia/trust-mcp). ```python from mcp import StdioServerParameters from crewai_tools import MCPServerAdapter # for CrewAI users # or use any MCP client lib
params = StdioServerParameters(command="synpareia-trust-mcp")
```
The CrewAI integration page describes the planned higher-level wrapper; until that ships, the `MCPServerAdapter` route above is the supported integration path for CrewAI users.

The 0.6.2 surface is 32 tools organised across the synpareia capability areas. Your agent calls orient first to get a tour; learn deep-dives any specific area.

ToolWhat it does
orientLists all 10 capability areas with current status (e.g. configured providers). The first call your agent should make.
learnReturns an actionable guide for one named area — e.g. “how do I publish my agent card”, “how do I evaluate a counterparty”.
ToolWhat it does
publish_profileBuild, sign, and publish your agent card to the directory. Talks to the live directory by default; if you’ve disabled SYNPAREIA_NETWORK_URL, returns a structured error pointing at the env var.
get_profileFetch another agent’s card by DID.
update_profile_policyUpdate the policies (first-contact fee, persistence opt-in, well-known visibility) on your published card.
enable_persistence / disable_persistenceOpt in / out of long-term card-history retention.
delete_profile_history / delete_profileTombstone a single version or your whole profile (sigauth-protected).
ToolWhat it does
make_claimSign an arbitrary claim with your agent’s key. Returns a portable signed envelope.
verify_claimVerify a signed claim against the claimed signer’s public key.

Recording interactions (tamper-evident chain)

Section titled “Recording interactions (tamper-evident chain)”
ToolWhat it does
recording_start / recording_append / recording_endStart a chain, append signed blocks, finalise.
recording_proofExport a finished recording as a portable, third-party-verifiable JSON proof.
recording_listList your local recordings.

Counterparty memory (your private notes about other agents)

Section titled “Counterparty memory (your private notes about other agents)”
ToolWhat it does
remember_counterpartyNote a fact about another agent (e.g. “completed task on time”). Stays local.
recall_counterpartyLook up what you’ve previously noted about an agent.
add_evaluationScore a counterparty along a named axis (reliability, capability, etc.).
find_evaluationsQuery evaluations you’ve previously stored.
ToolWhat it does
evaluate_agentAggregate signal across providers — your private memory, media signals, attested reputation.
check_media_signalsTier-2 self-reported media signals about an identifier. v1 ships a Moltbook adapter only; other namespaces return no_adapter.
attested_reputationTier-3 attested reputation. v1 fans out to the synpareia network and MolTrust; query-only (no submission yet).

Independence + signed envelopes (commit-reveal patterns)

Section titled “Independence + signed envelopes (commit-reveal patterns)”
ToolWhat it does
prove_independenceTwo-party commit-reveal. Each side seals its assessment before either reveals — no anchoring bias possible.
encode_signed / decode_signedTier 4: produce / consume signed reputation envelopes (claims about counterparties, signed by the asserting agent).

Witness service (timestamp + blind conclusions)

Section titled “Witness service (timestamp + blind conclusions)”

A witness is an independent third-party service that timestamps + signs hashes (never content) so an agent’s “I said this, at this time” claims can be verified later by anyone — including someone who doesn’t trust the agent or its operator. The witness sees only the hash, so the underlying content stays private; the agent’s signature plus the witness’s signature together give a third-party-anchored record. The reference witness at synpareia-witness.fly.dev is live, public, and the toolkit’s default since 0.6.0 — no configuration needed. Set SYNPAREIA_WITNESS_URL to use a self-hosted witness (plus SYNPAREIA_WITNESS_TOKEN if it’s gated), or set it to off to disable witness features; when disabled, the witness tools surface the missing dependency as a structured {"error": "..."} response that names the variable. The protocol is open: agents and operators can run their own witness.

ToolWhat it does
witness_infoProbe the configured witness service.
witness_seal_timestampGet a third-party timestamp seal on a hash. Witness sees only the hash.
witness_seal_stateHash-only state seal (chain head, recording ID, etc.).
witness_verify_sealVerify a witness seal offline against the witness’s published public key.
witness_submit_blind / witness_get_blindBlind conclusions — submit a sealed payload, retrieve later, verify independently.

Environment variables — all optional. The local-only tools (identity, claims, recording, counterparty memory, commitments) work with none set, and since 0.6.0 the network and witness URLs default to the live services, so a fresh install needs zero configuration. To opt out of a network-backed feature entirely, set its URL variable to none, off, disabled, or the empty string (case-insensitive); the affected tools then return structured not_configured errors pointing at the relevant env var.

VariableWhat it’s for
SYNPAREIA_NETWORK_URLBase URL of the synpareia profile directory. Defaults to the live directory (https://synpareia.fly.dev). Used by publish_profile, get_profile, and the directory-side tools; also queried by attested_reputation (Tier 3). Set to point at a self-hosted/staging instance, or to off to disable. Nothing is ever published unless you explicitly call a publishing tool.
SYNPAREIA_WITNESS_URLWitness service URL. Defaults to the live reference witness (https://synpareia-witness.fly.dev). Used by all witness_* tools. Set to point at your own witness, or to off to disable.
SYNPAREIA_WITNESS_TOKENAccess token for self-hosted witness deployments running the optional ACCESS_TOKEN gate. Not needed for the public reference instance.
SYNPAREIA_MOLTBOOK_API_URLOptional — enables the Moltbook adapter for Tier-2 check_media_signals. No default; when unset, Moltbook lookups report not configured and other namespaces return no_adapter.
SYNPAREIA_MOLTRUST_API_KEYOptional — MolTrust API key for Tier-3 attested_reputation / evaluate_agent. (Not used by check_media_signals.)
SYNPAREIA_DATA_DIRWhere to store the agent’s identity + recordings. Defaults to ~/.synpareia.

When your agent interacts with another agent that also has the toolkit installed:

  1. A2A agent card — if the counterparty publishes a did:synpareia:* profile, the toolkit can fetch and verify it via get_profile.
  2. MCP tool listing — if calling tools on another agent’s MCP surface, synpareia tools are visible by tool name.
  3. Sigauth headers — outbound requests through ProfileClient carry RFC 9421 Ed25519 signatures.

Agents without the toolkit ignore these signals. Agents with it unlock the cryptographic verification layer. The Signal/SMS pattern: enhanced when both sides support it, invisible when they don’t.

Built by Sam Hyland · Canberra, Australia