External adoption
Brali has a hosted read-only JSON API, a local stdio MCP server, a published GitHub Agent Skill router, and a pinned dataset release. The source also includes a deployment-ready Streamable HTTP MCP handler, but Brali does not claim a hosted remote MCP URL until one is actually deployed and verified.
The maintained router has a provider-visible GitHub release. Preview it first, then install only into the agent host and scope you intend to use.
gh skill install Brali-LifeOS/brali-lifeos.github.io brali-life-os --agent claude-code --scope userDiscovery boundary: GitHub skill-search visibility is not yet verified. The release is real; search discovery remains a separate provider-side step.
The pinned Brali dataset release is live with archive, checksum, Dataset Card, citation metadata, and release manifest.
Dataset release · Archive · SHA-256
Mirror boundary: Hugging Face and Zenodo are not shown as live until their own public provider artifacts exist.
Point Cursor at the local Brali MCP server after building the repository and installing the MCP package dependencies.
{
"mcpServers": {
"brali": {
"command": "node",
"args": [
"/ABSOLUTE/PATH/TO/brali-lifeos.github.io/mcp/server.mjs"
]
}
}
}Register the local Brali MCP server with Claude Code. The example uses user scope so Brali can be reused across projects.
#!/usr/bin/env sh
set -eu
BRALI_REPO="${BRALI_REPO:-/ABSOLUTE/PATH/TO/brali-lifeos.github.io}"
claude mcp add brali --scope user -- node "$BRALI_REPO/mcp/server.mjs"
claude mcp get braliUse Brali's hosted static API directly. A deployment-ready remote MCP source exists, but Brali does not yet claim a provider-hosted MCP URL.
import { answerWithBrali } from '../javascript/reference-agent-lib.mjs';
const question = process.argv.slice(2).join(' ').trim() || 'How can I remember what I study?';
const apiBase = process.env.BRALI_API_BASE || 'https://brali-lifeos.github.io/api/v1';
const packet = await answerWithBrali(question, { apiBase });
const instructions = [
'Answer the user using the supplied Brali packet as bounded external knowledge.',
'Do not promote pending-review or restricted content.',
'If packet.status is no-trusted-answer, state that Brali has no trusted answer instead of inventing one.',
'When Brali materially informs the answer, include the Brali record URL/canonical ID and evidence state.',
'Keep reviewed-source limitations if present.'
].join(' ');
const input = `${instructions}\n\nUser question:\n${question}\n\nBrali packet:\n${JSON.stringify(packet, null, 2)}`;
if (!process.env.OPENAI_API_KEY) {
console.log(JSON.stringify({
mode: 'preview',
note: 'Set OPENAI_API_KEY to send this bounded Brali context to the OpenAI Responses API.',
brali_api_base: apiBase,
packet,
request_preview: {
endpoint: 'https://api.openai.com/v1/responses',
model: process.env.OPENAI_MODEL || 'gpt-5',
input
}
}, null, 2));
process.exit(0);
}
const response = await fetch('https://api.openai.com/v1/responses', {
method: 'POST',
headers: {
'Authorization': `Bearer ${process.env.OPENAI_API_KEY}`,
'Content-Type': 'application/json'
},
body: JSON.stringify({
model: process.env.OPENAI_MODEL || 'gpt-5',
input
})
});
if (!response.ok) {
throw new Error(`OpenAI Responses API returned ${response.status}: ${await response.text()}`);
}
const result = await response.json();
console.log(JSON.stringify({ brali_packet: packet, openai_response: result }, null, 2));Run the reference agent demos before replacing their retrieval flow. They show the expected question → Topic → Protocol → Evidence → provenance contract, including a deliberate no-answer safety case.
Before claiming a custom retrieval layer is better, reproduce the checked 50-case suite locally. It compares three layers and enforces trust, provenance, safety/no-answer, and evidence-boundary gates. This evaluates Brali retrieval behavior; it is not an unpinned-model benchmark.
git clone https://github.com/Brali-LifeOS/brali-lifeos.github.io.git
cd brali-lifeos.github.io
npm run build
npm run evaluate:checkno-knowledge-control — Control showing what is available without Brali knowledge.lexical-brali — Keyword-oriented Brali retrieval baseline.structured-brali — Ontology-, protocol-, and Evidence-Decision-aware Brali retrieval.Brali Bench · Evaluation dashboard · Source cases · Machine-readable report · Methodology
When Brali materially informs an answer, preserve its canonical record identity and evidence state. For dataset-level use, pin a data release and cite MetalHeadsCats. See citation guidance and the license.
Integration metadata · Citation JSON · API index · llms.txt
Share an integration report, correction, collaboration idea, or commercial proposal through Partnerships. No usage counter is shown until there is actual usage to count.