Xynaptic MCP
# Xynaptic — Real-world data for AI agents
[](https://glama.ai/mcp/connectors/io.xynaptic.mcp/mcpxynaptic) [](https://glama.ai/mcp/servers/ericchengech/xynaptic-catalog)
**This repository also contains the Xynaptic MCP server implementation** — a zero-dependency Node.js MCP server (`index.js`, JSON-RPC 2.0 over stdio) that exposes this catalog to MCP clients, plus its `Dockerfile`. Run locally: `node index.js` (Node 18+). Hosted endpoint: mcp.xynaptic.io/mcp.
**220 pay-per-request data & AI APIs — full agent ecosystem: 19 domains + 3 agent skills, 10 AI vertical agents (OpenAI-compatible /v1/chat/completions with streaming), MCP server at mcp.xynaptic.io/mcp, travel & tourism (destination briefs, weather, routes, train trip planner), worldwide airport briefs, official FAA NOTAMs (airport & geospatial), invoice factoring risk (debtor & portfolio scoring), fiat↔USDC best-rail rates, pre-transaction wallet screening, crypto global metrics & movers (CMC), French legal watch, live French jobs (France Travail API) & recruitment tensions (BMO Dares), supplier compliance screening, used-car dealer check (anti-fraud before buying), the first x402-paid agent SKILLS (supplier due-diligence, payment fraud screening, crypto token audit — method + bundled data calls), battery-metals EV brief (lithium, copper, platinum, palladium), live French electricity demand & cross-border flows (RTE éCO2mix), ephemeral storage, watches, prepaid tickets. No API key. No mandatory subscription. Pay in USDC via [x402](https://x402.org) on Base, Solana or Polygon.**
```
GET https://api.xynaptic.io/v1/energy-price
- **Trust Layer**: machine-readable trust card at https://api.xynaptic.io/trust (free) — identity, x402 compliance, external probes (Market402 A 100/100, vet402 paid PASS), delivery evidence, and a live BTC cross-check CEX (Coinbase) vs DEX (DexScreener on-chain), refreshed on every call.
→ 402 Payment Required (price: $0.02)
→ sign EIP-3009 (or your x402 client)
→ retry with PAYMENT-SIGNATURE
→ 200 JSON: facts, assessment, summary, agent-ready actions
```
[](https://lumierepaycheck.org/e?url=https://api.xynaptic.io/v1/ai-energy)
**Machine-readable catalog (free):** https://api.xynaptic.io
**Live activity counter (free):** https://api.xynaptic.io/v1/popularity
## Monthly plans (no account, pay in USDC via x402)
Per-call is the default. A monthly plan replaces the per-call wallet dance with a
quota of **units** (`units = price_usd × 100`, so a $0.02 call costs 2 units).
Overage always falls back to pay-per-call x402 — a call is never blocked.
| Plan | Price | Scope | Quota |
|---|---|---|---|
| `llm` | $9/mo | 14 `ai-*` endpoints | 2,500 units · 100/h |
| `starter` | $15/mo | full catalog | 1,000 units · 200/h |
| `median` | $29/mo | full catalog | 3,000 units · 400/h |
| `pro` | $49/mo | full catalog | 8,000 units · 800/h |
| `devkit` | $15/mo | 6 dev endpoints | 1,000 units · 100/h |
| `security` — **Audit & Security** | $19/mo | **9 audit endpoints** (8 routes) | 1,000 units · 250/h |
| `svc-<service>` | $4/mo | **one service only** (16 AI Skills endpoints) | 570 units · 20–57/h |
The hourly ceiling is set per plan and must stay **at or above the cost of the most expensive
service in scope** (`security-sim` costs 100 units, hence 250/h on the security pack). A plan
that cannot serve its own priciest endpoint is a promise the API cannot keep.
`POST https://api.xynaptic.io/v1/sub-buy` with `{"plan":"security"}` returns the x402
challenge (19 USDC / 30 days), then the `sub_key` works as a Bearer token on every
in-scope endpoint (or `XYNAPTIC_SUB_KEY` for the MCP server).
### Per-service plans ($4/mo)
Every AI Skills endpoint except the composite `ai-brief-me` bundle can be subscribed to
**on its own** for $4/month, with the same `sub_key` mechanics:
`ai-chat` · `ai-search` · `ai-agent` · `ai-company` · `ai-finance` · `ai-energy` ·
`ai-health` · `ai-property` · `ai-insurance` · `ai-industry` · `ai-bank` ·
`ai-translate` · `ai-validate` · `skill-crypto-token-audit` ·
`skill-payment-fraud-screening` · `skill-supplier-due-diligence`
```bash
curl -X POST https://api.xynaptic.io/v1/sub-buy -d '{"plan":"svc-ai-chat"}'
# 402 challenge: 4 USDC / 30 days, then the response carries the sub_key
```
570 units per month, i.e. about **$5.70 of calls for $4** (−30 % vs pay-per-call):
285 `ai-chat` calls, 57 `ai-company` calls, 22 `skill-*` calls. Hourly caps follow the
same arithmetic (28/h for `ai-chat` and `ai-search`, 20/h for the other `ai-*`, 50/h for
the `skill-*`). The relay enforces the scope: any other endpoint returns
`403 endpoint_not_in_plan` and falls back to pay-per-call — never blocked. `ai-brief-me`
is a store bundle, it has no per-service plan.
**Audit & Security pack ($19/mo)** — the 9 audit endpoints only:
`mcp-audit` · `repo-security-audit` · `repo-security-audit-large` (same route, $2/call tier beyond 20 files) ·
`domain-ownership-proof` · `security-sim` · `wallet-watch` · `route-watch` ·
`route-watch-register` · `route-watch-history`. Equity/SEC `security-*` services stay pay-per-call.
**Standalone**: `wallet-watch` (address-poisoning watch, one payout wallet, hourly checks,
webhook alerts) is **$12/mo** — `POST /v1/wallet-watch/register`.
## Full service reference (new)
Every service, detailed per domain — **What it does** (description, sources) and **How to use** (route, parameters, price, payment flow): **[docs/services/](docs/services/README.md)** — **220 services, 19 domains**, same taxonomy as the live docs at https://xynaptic.io/docs.
> *If GitHub file pages return an error ("Error loading page" / 504), it is the GitHub viewer that is down, not the repo: the same tree is readable on https://xynaptic.io/docs and each card in raw form on https://raw.githubusercontent.com/ericchengech/xynaptic-catalog/main/docs/services/README.md*
## Domains
| Domain | Services | Highlights |
|---|---|---|
| ⚙️ Xynaptic Core | 7 | `ao` orchestrator · `store-put` · `ticket-buy` |
| 🤖 Skills | 17 | `ai-chat` OpenAI-compatible · `ai-agent` · 3 x402-paid SKILL.md |
| 📈 Crypto | 24 | `crypto-price` · `crypto-orderflow` · `crypto-validate` |
| 💵 Finance | 20 | `security-price` · `security-eu-price` · `security-analysis` |
| 🏦 Banking & Payments | 19 | `bank-statement-parse` · `wallet-screen` · `bank-sanctions-screen` |
| 🏢 Business | 18 | `company-kyb` · `company-financial` · `supplier-risk` |
| 💼 Jobs & Public | 16 | `job-search` · `tender-search` · `legal-brief` |
| 📰 News | 8 | `news-world` · `news-tech` · `news-eu` |
| ✈️ Flight | 9 | `flight-status` · `notam-brief` · `flight-route-risk` |
| 🚆 Rail | 8 | `rail-departures` · `rail-train-status` · `rail-disruptions` |
| 🗺 Travel & Tourism | 4 | `destination-brief` · `travel-weather` · `trip-planner` |
| ⚡ Energy | 11 | `energy-price` · `energy-charge-window` · `energy-demand` |
| 🏭 Industry & Commodities | 11 | `commodity-price` · `industrial-supply` · `supplier-compliance` |
| 🏠 Property | 11 | `property-valuation` · `property-price` · `property-risk` |
| 🛡️ Insurance | 10 | `insurance-policy` · `insurance-profile` · `insurance-eligibility` |
| 🏥 Health | 8 | `health-drug-info` · `health-clinical-trials` · `health-news` |
| 🛠️ Dev Kit | 6 | `code-review` · `code-fix` · `code-refactor` · `test-generate` · `error-explain` · `security-review` |
| 🛡️ Audit & Security | 9 | `mcp-audit` · `repo-security-audit` · `domain-ownership-proof` · `security-sim` · `wallet-watch` · `route-watch` |
| 🍽️ Leisure & going out | 4 | `restaurant-top` · `restaurant-detail` · `restaurant-search` · `restaurant-nearby` |
## Dev Kit — LLM-powered developer tools
Six developer tools that run on flash-class frontier models (GLM-5.3-Flash / DeepSeek-V4.1-Flash auto-routing, JSON-strict output) — fast, cheap, expert-grade on bounded tasks. Paste code or a stacktrace, get expert-grade analysis back. Pay per call in USDC, like every catalog service.
| Tool | What it does | Input | Price |
|---|---|---|---|
| `code-review` | Severity-ranked code review with quality score; optional `include_fix` for suggested fixes | `code` (10–12000 chars, any language) | $0.05 |
| `code-fix` | Buggy code in, corrected version out — with the bug explained line by line | `code` | $0.05 |
| `code-refactor` | Behavior-preserving refactor: cleaner structure, readability wins, recommended tests | `code` | $0.05 |
| `test-generate` | Functional test cases, code skeletons and security checks for any snippet | `code` | $0.05 |
| `error-explain` | Paste a stacktrace, get the root cause explained with a concrete fix | `error` (stacktrace or error text) | $0.03 |
| `security-review` | Security audit of code or config: vulnerabilities, secrets exposure, injection risks, hardening advice | `content` (code or config, multi-format) | $0.05 |
**Dev Kit subscription — $15/mo for coding agents.** 1,000 units/month (100/hour cap) covering all 6 dev endpoints. Everything else stays pay-per-call.
```
POST https://api.xynaptic.io/v1/sub-buy
{ "plan": "devkit" }
-> 402 x402 $15 USDC -> sign -> sub_key
Authorization: Bearer *** on the 6 dev endpoints
```
```
POST https://api.xynaptic.io/v1/code-review
-> 402 Payment Required (price: $0.05)
-> pay via x402, retry with PAYMENT-SIGNATURE
-> 200 JSON: { findings: [...], quality_score, ... }
```
All six are live and QA-settled on Base, Solana and Polygon. Also exposed through the MCP server (mcp.xynaptic.io/mcp).
## Agent Skills — the first x402-paid skills
A skill is a complete method (SKILL.md, Claude/OpenClaw-compatible) that chains several catalog services with decision grids, thresholds and abort rules. Buy the method once ($0.25), then execute it: each data call is paid per-use via x402 as usual.
| Skill | What it does | Chains | Price |
|---|---|---|---|
| `skill-supplier-due-diligence` | Vet a French supplier before contracting: 5-step method (identity, KYB, compliance, financials, press) with GO / GO-WITH-CONDITIONS / NO-GO grid | company-search, company-kyb, supplier-compliance, company-financial, company-news | $0.25 |
| `skill-payment-fraud-screening` | Screen any outgoing payment (fiat or crypto) before sending: sanctions signals, wallet screening, structure risk, with SEND / SEND-WITH-CHECKS / BLOCK rules | bank-payment-risk, bank-sanctions-screen, wallet-screen | $0.25 |
| `skill-crypto-token-audit` | Audit a token before buying: contract security, tokenomics, market state, flow — BUY / CAUTION / AVOID grid | crypto-validate, crypto-tokenomics, crypto-price, crypto-risk, crypto-orderflow | $0.25 |
```
GET https://api.xynaptic.io/v1/skill-supplier-due-diligence
-> 402 Payment Required (price: $0.25)
-> pay via x402, retry with PAYMENT-SIGNATURE
-> 200 JSON: { skill_md: "<full SKILL.md — method, grids, exact endpoint calls>" }
```
The returned `skill_md` is a standard SKILL.md (YAML frontmatter + step-by-step method): an agent can execute it directly, calling the chained services as it goes.
## The AI layer (new)
```
INFERENCE
POST /v1/ai/chat — $0.02
OpenAI-compatible {model, messages[]}
→ x402 → glm-5.3-flash (fallback: deepseek)
→ JSON answer + token usage
ORCHESTRATION
POST /v1/ai/agent — $0.05
{task, siren?} → ONE payment
→ internal tools (registry data — free)
→ LLM synthesis → structured report
VERTICAL AGENTS
POST /v1/ai/company — $0.10
{siren, focus?: risk|financial|overview}
→ 5 registry tools (deep dossier)
→ LLM KYB specialist: identity,
management, financials, risk signals,
VERDICT (proceed/review/stop) + confidence
POST /v1/ai/finance — $0.05
{asset, question?}
→ 5 market tools (price, orderflow,
derivatives, risk)
→ LLM analyst: sourced report
(Kraken, Hyperliquid), no financial advice
```
The client agent pays once; Xynaptic monetizes data + orchestration + inference. The LLM uses ONLY tool data — empty context = honest "no data" instead of invention (verified in QA twice).
```
POST /v1/ai/chat — $0.02 per call
OpenAI-compatible {model, messages[]}
→ x402 payment → glm-5.3-flash
(fallback: deepseek-4.1-flash)
→ JSON answer + token usage
POST /v1/ai/agent — $0.05 per task
{task, siren?}
→ ONE x402 payment
→ internal tools (company registry,
financials, officers, KYB signals)
— free for us, data-true
→ LLM synthesis (cites the numbers)
→ structured report
```
The agent client pays once; Xynaptic monetizes data + orchestration + inference. The LLM is instructed to use ONLY the tool data — if the context is empty, it says so instead of inventing (verified in QA).
## The agent ecosystem (new)
Six bricks that make an agent **live** in Xynaptic — not just query it:
```
STORE deposit any JSON, encrypted at rest, claim_secret = the only key
TTL 1-30 days, then cryptographically destroyed. Handoff-friendly:
share {store_id, claim_secret} to pass an object to another agent.
WATCH the daycare: "tell me when BTC < 80000" or "alert if company
552032534 changes" — our evaluator checks hourly (cron), result
is encrypted with your watch_secret, delivered via watch-status.
TRANSLATE bring your own data: paste raw text (CSV, email, notice, listing)
→ clean structured JSON matching your target_schema.
BRIEF-ME the shopping basket: ONE payment → up to 3 companies + 5 assets +
2 news regions, collected internally + LLM morning brief synthesis.
VALIDATE the devil's advocate: submit a thesis, we stress-test it against
real data (registry, market) — strengths, weaknesses, factual
contradictions with cited figures, counter-thesis, verdict + confidence.
TICKETS prepaid credit, no account: pay once (0.5-20 USDC), then send
header `X-TICKET: ticket_id|ticket_secret` on ANY service —
20% off, no per-call blockchain payment, valid 90 days.
Balance check: POST /v1/ticket-status (free).
```
Each brick is QA-proven with real on-chain settlement, encrypted at rest
where it stores anything, and expires honestly (TTL destroys, no dark patterns).
## The resident domains (new)
Full domains so an agent never has to leave Xynaptic:
- **BANKING/MONETICS (16 services)** — from raw statement parsing to fraud detection,
chargeback classification, counterparty KYB and payment routing: the complete
journey of a banking agent. Bring-your-own-data, nothing stored, disclaimers on
every decision-adjacent route.
- **SECURITIES (19 services)** — the full services on official US regulator data (EDGAR filings, XBRL fundamentals, insider, earnings, dividends) + 6 EU routes (Yahoo Finance EU markets): live prices Paris/XETRA/LSE/Amsterdam, technicals RSI/momentum, analysis, compare, risk — local currency + EUR
(SEC EDGAR) + live prices: identity, fundamentals, technicals, risk, filings
full-text, insider transactions, dividends, earnings, daycare watches,
devil's-advocate validation and portfolio analysis.
- **HEALTH + REAL ESTATE + INDUSTRY + INSURANCE (42 services)** — four new domains, all QA-settled: official health sources (openFDA, ClinicalTrials.gov, PubMed, ScopeSanté, RPPS 2.3M pros), official French real-estate transactions (DGFiP DVF, 915k+ sales, median €/m²), live commodities + INSEE producer prices + supplier intelligence, and the AI insurance suite (policy parsing, claim structuring, coverage verdicts, legal-basis quotes).
- **WEB + QUANT + FRENCH PUBLIC DATA (11 services)** — `ai-search` gives agents real web access
with sourced answers; the crypto quant pack (portfolio, historical, correlation, screener, gas fees)
completes the trading stack; `company-news` closes the KYB loop with press; `intermodal-compare`
crosses real SNCF data with flight estimates; `tender-search` opens official public procurement;
`gas-tariff-zone` and `news-economy-fr` add CRE-regulated data and live economy press.
- **LEISURE & GOING OUT (4 services)** — `restaurant-top`, `restaurant-detail`, `restaurant-search`, `restaurant-nearby`: 139 restaurants across 88 cities, Michelin distinctions from Wikipedia/Wikidata, official hygiene checks (Alim Confiance, FSA, NYC DOHMH), contacts and booking links, opening hours from OpenStreetMap when they exist and `null` otherwise, plus the volume layer `restaurant-nearby`: every food place OSM knows in a city zone (restaurant, fast food, bar, pub, cafe) with distance from your point and `open_now` only when the published hours allow a certain answer. No editorial score on that layer. Sources cited in every record.
- **INTERNATIONAL COMPANIES (2 services)** — GLEIF LEI registry, 200+ jurisdictions
worldwide: search by name, full identity by LEI. French registry (11 services)
remains the deep end; LEI is the global layer.
## The company services
```
SEARCH → PROFILE → PEOPLE / NETWORK / FINANCIAL
│
RELATIONSHIP (two companies)
│
RELATIONSHIP / COMPARE (two companies)
│
KYB — the $0.10 dossier
(identity, officers, accounts, risk,
VAT, verified sources, agent verdict)
│
AI AGENT DECISION
```
Built on the French state registry (RNE) — open data, no key required. Danone example: "minimal risk (0) — active, 13 officers, 2025 accounts, VAT present. Action: proceed."
## The crypto services
```
MARKET / NEWS / ONCHAIN
│
ORDERFLOW ──── DERIVATIVES ──── DEFI / TOKENOMICS
│
TRADING ANALYSIS
│
DEEP TRADING ANALYSIS
│
CRYPTO VALIDATE ──→ AI AGENT
```
**New — the deep read is now fully wired:** `crypto-deep-trading-analysis` synthesizes 5 dimensions in one conviction score — multi-timeframe trend, volatility regime, news sentiment, orderflow (CVD from the live tape) and derivatives (funding, OI, basis). Example: "BTCUSD: neutral (-0.2) — 4h bearish but flow 0.96, funding 0.11%/yr".
Every response chains to the next API you might need (`agent.related_xynaptic_apis`): buy one, discover the rest.
## Try it free — sandbox and Xyna
Two free entry points let a visitor or an agent test the catalog before paying anything.
**Sandbox — `GET /v1/demo/{service}`**
Three services are open: `crypto-price`, `energy-price`, `company-search`.
```
curl https://api.xynaptic.io/v1/demo/crypto-price
curl "https://api.xynaptic.io/v1/demo/company-search?q=michelin"
```
- 10 calls per IP per day (`X-Demo-Remaining` header, `demo_meta` block in the JSON)
- identical payload to the paid endpoint, real upstream data
- HTTP 429 once the quota is spent, HTTP 404 for any other service
**Guide — `POST /v1/ask` (Xyna)**
Xyna reads the catalog and answers in natural language with one to three matching
services and their prices. Stateless, no account, English by default (answers in the
language of the question).
```
curl -X POST https://api.xynaptic.io/v1/ask \
-H 'content-type: application/json' \
-d '{"question":"I need to check a French company before signing"}'
```
```json
{"xyna":"Xyna, the Xynaptic guide","free":true,"calls_today":1,"calls_limit":20,
"answer":"Start with company-search ($0.02) to resolve the company by name or SIREN, then ...",
"suggestions":[{"service":"company-search","price_usd":0.02,"demo":true}, ...]}
```
- 20 questions per IP per day (`X-Ask-Remaining` header)
- the model runs on local inference: no third-party API is exposed, no key is needed
- scope is the Xynaptic catalog only; off-topic questions are refused
Both are also wired into the site: a **Try it — free** button on the matching endpoint
cards in the docs, and a **Xyna** chat bubble on every page.
## Quick start
```bash
# Discover the full catalog (free, machine-readable)
curl https://api.xynaptic.io/
# See live activity
curl https://api.xynaptic.io/v1/popularity
# Company due diligence (the premium dossier):
curl https://api.xynaptic.io/v1/company-kyb?siren=552032534
# Any x402 SDK pays automatically, e.g. TypeScript:
# const data = await client.fetch("https://api.xynaptic.io/v1/crypto-validate?address=0x...")
```
Or search us on the **x402 Bazaar**: `searchX402Resources({ query: "orderflow" })`
## Response schema (same across all services)
```json
{
"service": "crypto-orderflow",
"facts": { "cvd": 5.05, "buy_sell_ratio": 0.81, "whale_trades": [] },
"assessment": { "bias": "bullish_flow" },
"summary": "BTCUSD tape: buy pressure (ratio 0.81), CVD +5.05.",
"agent": { "actionability": "high", "recommended_actions": ["consider_long"],
"related_xynaptic_apis": [{ "endpoint": "/v1/crypto-derivatives" }] },
"data": { "source": "Kraken live tape", "freshness": "real-time" }
}
```
## Networks
- Base (eip155:8453) — USDC
- Polygon (eip155:137) — USDC
- Solana — USDC
*Real-time sources: Kraken, Hyperliquid, DefiLlama, CoinGecko, SNCF/Navitia, aviationweather, EPEX/awattar, Open-Meteo, honeypot.is, DexScreener. Curated 2h digests for news. Built by an independent builder.*TDQS
Scored across 24 tools
Most tools map to clearly distinct domains or resources (French company data, energy, crypto, travel, legal, tenders), and descriptions explicitly differentiate sibling tools such as crypto-price vs crypto-global vs crypto-movers. The main ambiguity is xynaptic-catalog acting as a catch-all alongside the flagship tools, plus several 'price' tools, but the descriptions largely mitigate this.
All names use readable kebab-case and are domain-prefixed, but there is no consistent verb_noun pattern: most are noun compounds (company-profile, crypto-price), while others are verb-final (ticket-buy, bank-transaction-categorize) or utility names (xynaptic-status, xynaptic-catalog). The convention is readable but not fully predictable.
24 tools is at the heavy end of the 16-25 range and creates meaningful selection load for an agent. The broad multi-domain marketplace purpose justifies many flagship tools, and xynaptic-catalog consolidates the 200+ long tail, but the direct surface still feels crowded.
Core domains are represented and the catalog explicitly fills long-tail gaps such as bank analytics, regional news, EU stocks, property rents, and notam-area. However, several flagship domains are shallow (bank only categorizes, property only sale-price benchmarks, news only global), so agents must use the catalog for common adjacent operations.