Turn JSON bytes into one stable Rust-produced encoding, then validate typed host documents. A canonical digest identifies bytes; it does not authenticate their author or authorize an effect.
When to use this
Use this when data must have the same encoding and digest across Python, TypeScript and Rust.
Install the matching API preview
These examples use the API 2 preview, not names promised by the public RC5 release. Existing RC5 entry points remain available. The Python wheel below is verified on Windows x64; it is not a cross-platform release.
python -m pip install https://docs.soterias.io/downloads/api-preview/soterias_local-0.1.0rc5%2Bapi2-cp311-abi3-win_amd64.whlnpm install https://docs.soterias.io/downloads/api-preview/soterias-core-0.1.0-rc.5-api.2.tgz
npm install --save-dev typescript@5.8.3 @types/node@24Python imports soterias. TypeScript imports @soterias/core. Use Python 3.11+ for this wheel and Node 22.23.4+ for these examples. For a .mts file, compile with npx tsc example.mts --target ES2022 --module NodeNext --moduleResolution NodeNext --skipLibCheck, then run the emitted .mjs file.
Before you run
This program creates its own local example state. After installing the matching SDK, save the chunks in order and run python example.py or compile and run example.mts. No credentials, model or service are required.
Application walkthrough
Choose Python or TypeScript. Create example.py or example.mts and append that language's chunks in order. Each language forms one complete program; indentation and open blocks carry across chunks.
Import the packages
Import the SDK and standard-library helpers. These are the real package exports used by the executable program.
import json
from soterias import canonicalize, canonical_digest, InputErrorimport assert from 'node:assert/strict';
import { canonicalize } from '@soterias/core';Produce stable bytes using Rust canonical JSON
Property order and insignificant whitespace do not change the encoding.
# %% Produce stable bytes using Rust canonical JSON
# Property order and insignificant whitespace do not change the encoding.
first = b'{ "invoice": 42, "currency": "CHF" }'
second = b'{"currency":"CHF","invoice":42}'
encoded = canonicalize(first)
assert encoded == canonicalize(second)
assert canonical_digest(first) == canonical_digest(second)// %% Produce stable bytes using Rust canonical JSON
// Property order and insignificant whitespace do not change the encoding.
const first = Buffer.from('{ "invoice": 42, "currency": "CHF" }');
const second = Buffer.from('{"currency":"CHF","invoice":42}');
const encoded = canonicalize(first);
const same = canonicalize(second);
assert.deepEqual(encoded.canonical, same.canonical);
assert.equal(encoded.sha256, same.sha256);Reject ambiguous input at the boundary
Never accept duplicate JSON keys by silently choosing one value.
# %% Reject ambiguous input at the boundary
# Never accept duplicate JSON keys by silently choosing one value.
try:
canonicalize(b'{"invoice":42,"invoice":99}')
except InputError:
pass
else:
raise AssertionError("Duplicate keys must be rejected")// %% Reject ambiguous input at the boundary
// Never accept duplicate JSON keys by silently choosing one value.
assert.throws(() => canonicalize(Buffer.from('{"invoice":42,"invoice":99}')));Pass the stable bytes to your application
A digest identifies data. It does not establish who signed it.
# %% Pass the stable bytes to your application
# A digest identifies data. It does not establish who signed it.
print(json.dumps({"status": "PASS", "canonical": encoded.decode(),
"digest": canonical_digest(first)}))// %% Pass the stable bytes to your application
// A digest identifies data. It does not establish who signed it.
console.log(JSON.stringify({ status: 'PASS', canonical: encoded.canonical.toString(),
digest: encoded.sha256 }));MCP composition
This complete stdio server calls this feature's real SDK operation. Its bounded tool inputs cannot replace host-selected trust, checkpoint or storage paths. Save it as mcp.mjs and configure your MCP client to launch node mcp.mjs. Public data inputs use the same environment settings as the application program. The code is a custom application tool, not a newly claimed built-in Soterias endpoint.
Install the matching adapter with npm install https://docs.soterias.io/downloads/api-preview/soterias-mcp-server-0.1.0-rc.5-api.2.tgz. Its SDK peer dependency shares the same native class identities as your installed preview.
mcp.mjs
Create mcp.mjs and append these chunks in order. Together they form one complete program; indentation carries across chunks.
Step 1
Import the SDK and standard-library helpers. These are the real package exports used by the executable program.
import { Soterias, observe, canonicalize, Metrics, audit, HistoricalPolicyInputs } from '@soterias/core';
import fs from 'node:fs';
import os from 'node:os';
import path from 'node:path';
import { McpServer } from '@modelcontextprotocol/server';
import { serveStdio } from '@modelcontextprotocol/server/stdio';
import * as z from 'zod/v4';
// Trusted startup: paths, trust and credentials are fixed by the host.
const root = fs.mkdtempSync(path.join(os.tmpdir(), 'soterias-mcp-'));
const sdk = new Soterias({workflow: 'invoice-tools', directory: path.join(root, 'metrics')});
const sum = input => input.cents.reduce((a, b) => a + b, 0);
// Agent input is bounded data. It cannot replace host configuration or trust.
const handle = serveStdio(() => {
const server = new McpServer({name: 'soterias-canonical', version: '1.0.0'});
server.registerTool('canonical_json', {
description: 'Use canonical through the host-selected native SDK.',
inputSchema: z.object({json: z.string().min(1).max(65536)}).strict(),
}, async input => {
try {
const result = await (async () => {
const result = canonicalize(Buffer.from(input.json));
return {canonical: result.canonical.toString(), sha256: result.sha256};Step 2
Finish outstanding work and release owned resources. Cleanup errors must not silently replace an existing application error.
})();
return {content: [{type: 'text', text: JSON.stringify(result, (_key, value) => typeof value === 'bigint' ? value.toString() : value)}]};
} catch {
return {isError: true, content: [{type: 'text', text: 'feature_request_refused'}]};
}
});
return server;
}, {maxSubscriptions: 0});
process.stdin.once('end', () => {
void (async () => { await handle.close(); await sdk.closeAsync(); })();
});Test the tool over a real MCP connection
Save this client next to mcp.mjs, then run node client.mjs. It starts the host, calls the actual feature tool and closes the connection. It does more than list the tool catalog.
client.mjs
Create client.mjs and append these chunks in order. Together they form one complete program; indentation carries across chunks.
Step 1
Import the SDK and standard-library helpers. These are the real package exports used by the executable program.
import { Client } from '@modelcontextprotocol/client';
import { StdioClientTransport } from '@modelcontextprotocol/client/stdio';
import { fileURLToPath } from 'node:url';
const client = new Client({name: 'feature-example', version: '1.0.0'});
try {
await client.connect(new StdioClientTransport({
command: process.execPath,
args: [fileURLToPath(new URL('./mcp.mjs', import.meta.url))],
env: Object.fromEntries(Object.entries(process.env).filter(([, value]) => value !== undefined)),
}));
const result = await client.callTool({name: 'canonical_json', arguments: {json: '{"invoice":42,"currency":"CHF"}'}});
if (result.isError) throw new Error('Feature request refused');
console.log(JSON.stringify(result));
} finally { await client.close(); }Check the result
The application program prints a JSON result and preserves native refusals. For protected operations inspect status and target_observed: transport success does not imply a completed effect. The pure examples contain assertions for their exact expected values. Successful startup is not proof of production enrollment, a physical human review or current authority.
Put it into a workflow
Use the AML walkthrough for protected document reads, reviewed writes and signed outcome checks. Use multi-agent workflows for personal and enterprise orchestration. Keep each agent's proposals separate from operator authority.