MCP self-orientation: /context + /verify + architecture resources

Any agent (Claude Code via MCP stdio, or sub-agents via HTTP :3700)
can now self-orient without human explanation:

GET /context returns:
  - System purpose and name
  - All datasets with row counts
  - All vector indexes with backends
  - Available models and their strengths
  - Complete tool list with rules
  - Current VRAM state

POST /verify fact-checks any claim about a worker against the golden
data. Agent says "worker 1313 is a Forklift Operator in IL with
reliability 0.82" → endpoint returns verified=true/false with exact
discrepancies.

MCP resources (stdio path for Claude Code):
  - lakehouse://system — live system status
  - lakehouse://architecture — full PRD
  - lakehouse://instructions — agent operating manual
  - lakehouse://playbooks — successful operations database
  - lakehouse://datasets — dataset listing

This is the "command and control" layer J asked for: any agent
connecting to this system gets the context it needs to operate
independently. No human intermediary required.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
This commit is contained in:
root 2026-04-17 00:41:46 -05:00
parent 67ab6e4bac
commit a001a21902
2 changed files with 132 additions and 3 deletions

View File

@ -4,7 +4,8 @@
"command": "bun",
"args": ["run", "/home/profit/lakehouse/mcp-server/index.ts"],
"env": {
"LAKEHOUSE_URL": "http://localhost:3100"
"LAKEHOUSE_URL": "http://localhost:3100",
"MCP_TRANSPORT": "stdio"
}
}
}

View File

@ -172,7 +172,75 @@ server.tool(
},
);
// Resources
// Resources — these give any MCP client full context about the system
server.resource("lakehouse://system", "lakehouse://system", async (uri) => {
const health = await api("GET", "/health");
const datasets = await api("GET", "/catalog/datasets") as any[];
const indexes = await api("GET", "/vectors/indexes") as any[];
const vram = await api("GET", "/ai/vram");
const agent = await api("GET", "/vectors/agent/status");
const buckets = await api("GET", "/storage/buckets");
const text = `# Lakehouse System Status
## Health: ${health === "lakehouse ok" ? "OK" : JSON.stringify(health)}
## Datasets (${datasets.length})
${datasets.map((d: any) => `- ${d.name}: ${d.row_count || "?"} rows`).join("\n")}
## Vector Indexes (${indexes.length})
${(indexes as any[]).map((i: any) => `- ${i.index_name}: ${i.chunk_count} chunks (${i.vector_backend || "parquet"})`).join("\n")}
## GPU
- Used: ${vram?.gpu?.used_mib || "?"}/${vram?.gpu?.total_mib || "?"} MiB
- Models loaded: ${(vram?.ollama_loaded || []).map((m: any) => m.name).join(", ") || "none"}
## Autotune Agent
- Running: ${agent?.running}, Trials: ${agent?.trials_run}, Promotions: ${agent?.promotions}
## Buckets (${(buckets as any[])?.length || 0})
${(buckets as any[] || []).map((b: any) => `- ${b.name}: ${b.backend} (${b.reachable ? "reachable" : "DOWN"})`).join("\n")}
## Services
- Lakehouse Gateway: :3100
- AI Sidecar: :3200
- Agent Gateway: :3700
- Langfuse: :3001
- MinIO S3: :9000
- Ollama: :11434
## Available Models
- qwen3: 8.2B, 40K context, thinking+tools (best for reasoning)
- qwen2.5: 7B, 8K context (best for fast SQL generation)
- mistral: 7B, 8K context (general generation)
- nomic-embed-text: 137M (embedding, automatic)
`;
return { contents: [{ uri: uri.href, mimeType: "text/plain", text }] };
});
server.resource("lakehouse://architecture", "lakehouse://architecture", async (uri) => {
// Read the PRD directly
const prd = await Bun.file("/home/profit/lakehouse/docs/PRD.md").text().catch(() => "PRD not found");
return { contents: [{ uri: uri.href, mimeType: "text/markdown", text: prd }] };
});
server.resource("lakehouse://instructions", "lakehouse://instructions", async (uri) => {
const instructions = await Bun.file("/home/profit/lakehouse/mcp-server/AGENT_INSTRUCTIONS.md").text().catch(() => "Instructions not found");
return { contents: [{ uri: uri.href, mimeType: "text/markdown", text: instructions }] };
});
server.resource("lakehouse://playbooks", "lakehouse://playbooks", async (uri) => {
const r = await api("POST", "/query/sql", {
sql: "SELECT * FROM successful_playbooks ORDER BY timestamp DESC LIMIT 20"
});
const rows = r?.rows || [];
const text = rows.length === 0
? "No playbooks yet. Log successful operations with the log_success tool."
: rows.map((p: any) => `## ${p.operation}\n- Approach: ${p.approach}\n- Result: ${p.result}\n- Context: ${p.context || "—"}\n`).join("\n");
return { contents: [{ uri: uri.href, mimeType: "text/markdown", text: `# Successful Playbooks\n\n${text}` }] };
});
server.resource("lakehouse://datasets", "lakehouse://datasets", async (uri) => {
const r = await api("GET", "/catalog/datasets") as any[];
const text = r.map(d => `${d.name}: ${d.row_count || "?"} rows`).join("\n");
@ -204,7 +272,67 @@ async function main() {
try {
// Health
if (url.pathname === "/health") return ok({ status: "ok", lakehouse: BASE, tools: 9 });
if (url.pathname === "/health") return ok({ status: "ok", lakehouse: BASE, tools: 11 });
// Self-orientation: any agent calls this first to understand the system
if (url.pathname === "/context") {
const instructions = await Bun.file("/home/profit/lakehouse/mcp-server/AGENT_INSTRUCTIONS.md").text().catch(() => "");
const datasets = await api("GET", "/catalog/datasets") as any[];
const indexes = await api("GET", "/vectors/indexes") as any[];
const vram = await api("GET", "/ai/vram");
return ok({
system: "Lakehouse Staffing Co-Pilot",
purpose: "AI anticipates staffing coordinator needs — pre-matches workers to contracts, surfaces alerts, builds playbooks from successful operations",
instructions: instructions.slice(0, 3000),
datasets: (datasets || []).map((d: any) => ({ name: d.name, rows: d.row_count })),
indexes: (indexes || []).map((i: any) => ({ name: i.index_name, chunks: i.chunk_count, backend: i.vector_backend })),
models: { qwen3: "8.2B reasoning+tools", qwen2_5: "7B fast SQL", mistral: "7B generation", nomic: "137M embedding" },
vram: vram?.gpu,
tools: ["/search","/sql","/match","/worker/:id","/ask","/log","/playbooks","/profile/:id","/vram","/context","/verify"],
rules: [
"Never hallucinate — only state facts from tool responses",
"SQL for counts/aggregations, hybrid /search for matching",
"Log every successful operation to /log",
"Check /playbooks before complex tasks",
"Verify worker details via /worker/:id before communicating",
],
});
}
// Verification endpoint — agent can check any claim against SQL
if (url.pathname === "/verify") {
const b = await json();
// b.claim: "worker 4925 is a Forklift Operator in IL with reliability 0.82"
// b.worker_id: 4925
// b.checks: { role: "Forklift Operator", state: "IL", reliability: 0.82 }
if (!b.worker_id) return err("worker_id required");
const r = await api("POST", "/query/sql", {
sql: `SELECT * FROM ethereal_workers WHERE worker_id = ${b.worker_id}`
});
const worker = r?.rows?.[0];
if (!worker) return ok({ verified: false, reason: `worker ${b.worker_id} not found` });
const checks = b.checks || {};
const failures: string[] = [];
for (const [field, expected] of Object.entries(checks)) {
const actual = worker[field];
if (actual === undefined) continue;
if (typeof expected === "number") {
if (Math.abs(Number(actual) - expected) > 0.05) {
failures.push(`${field}: claimed=${expected} actual=${actual}`);
}
} else if (String(actual).toLowerCase() !== String(expected).toLowerCase()) {
failures.push(`${field}: claimed=${expected} actual=${actual}`);
}
}
return ok({
verified: failures.length === 0,
worker_id: b.worker_id,
worker_name: worker.name,
failures,
actual: worker,
});
}
// Tool: hybrid search
if (url.pathname === "/search") {