@nymrel/mcp-hub
OfficialSummary: This MCP server provides Nymrel-oriented tools, resources, and prompts for AI-commerce audits, agent safety, cryptographic proofs, crawling, multi-agent coordination, trust artifacts, quoting, local-model routing, commerce actions, sandboxing, and human-approval cards.
Audit URLs or HTML for AI commerce readiness and structured-data checks.
Inspect shell commands for destructive or path-traversal risks.
Claim, renew, release, and inspect repo/agent leases with fencing and studio roles.
Generate Schema.org JSON-LD,
/llms.txt, androbots.txtmachine-trust artifacts.Create and verify Merkle execution receipts and signatures.
Scrape/crawl public pages and convert HTML to Markdown with metadata extraction.
Emit agent heartbeats and query fleet status.
Calculate service quotes with scope, preset, currency, rush, and feature inputs.
Probe local GPU/Ollama status and route tasks across local/balanced/frontier tiers.
Run UCP quote, negotiate, commit, and x402 micropayment actions.
Mask secrets, inspect isolation, and enforce token-spend guardrails.
Render A2UI decision/review cards in Warm Paper styling.
Dispatch structured inter-agent bus messages (handoffs, reviews, broadcasts).
Access resources
nymrel://status,nymrel://ecosystem, andnymrel://llms-manifest.Use prompt templates for UCP audits, command safety, and two-seat mission setup.
Support modern MCP
server/discoverand initialize-era protocol compatibility over stdio.Note: the README describes a 15th web-search tool, but the supplied schema lists 14 tools and does not include it.
Enables agents to probe local GPU/Ollama status, calculate token dollar savings, and route tasks across Luna/Terra/Sol tiers for local model inference.
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@@nymrel/mcp-hub Audit https://example.com for commerce readiness and show the score."
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
Nymrel MCP Hub
Choose your starting point
People using Nymrel through a connected assistant: ask for a specific outcome in your existing conversation. An installed package alone does not establish a working connection; your host must support and enable this server. The example simulators are development previews rather than a hosted customer app.
Developers configuring an MCP client: use the source setup, build once, and run
--list-tools,--list-resources, or--list-promptsto inspect a catalog without starting a long-running server. Configure your client with an absolute path and--stdio.Agents: use the protocol contract for your negotiated MCP revision and the discovered input schemas. Keep protocol stdout free of progress text, inspect errors and structured results, and retain source/version context when handing an outcome back to a person.
A dual-engine Model Context Protocol server exposing 15 Nymrel tools, 3 resources, and 3 prompt templates through TypeScript/Node.js and Python stdio entry points.
The repository supports the modern MCP 2026-07-28 request model while retaining initialize-era compatibility through 2025-11-25. It does not currently claim a hosted transport or verified npm/PyPI publication.
Related MCP server: wham-engine
Distribution status
Registry publication is unverified. As of September 1, 2026, this repository contains npm and Python package manifests and gated release workflows, but it does not contain a registry receipt proving that @nymrel/mcp-hub or nymrel-mcp-hub is publicly installable.
Until an npm or PyPI receipt is independently verified:
do not use
npx @nymrel/mcp-hubas a documented installation path;do not use
pip install nymrel-mcp-hubas a documented installation path;run the server from a pinned source checkout using the instructions below;
treat a local build, GitHub artifact, tag, or workflow result as distinct from registry publication.
Publishing, releasing, tagging, and trusted-publisher configuration are intentionally outside this documentation correction.
Run from source
TypeScript / Node.js
Requirements: Node.js 22 or 24 and Corepack.
git clone https://github.com/nymrel/nymrel-mcp-hub.git
cd nymrel-mcp-hub
corepack npm@12.0.2 ci
corepack npm@12.0.2 run build
node ./bin/mcp-server.js --help
node ./bin/mcp-server.js --stdioPoint an MCP client at the built source checkout with an absolute path:
{
"mcpServers": {
"nymrel": {
"command": "node",
"args": [
"/absolute/path/to/nymrel-mcp-hub/bin/mcp-server.js",
"--stdio"
]
}
}
}The Node entry point imports compiled files from dist/, so run the build before starting the server.
Python
Requirements: Python 3.11 through 3.14.
git clone https://github.com/nymrel/nymrel-mcp-hub.git
cd nymrel-mcp-hub
python -m venv .venv
# macOS/Linux: source .venv/bin/activate
# Windows PowerShell: .venv\Scripts\Activate.ps1
python -m pip install --editable .
nymrel-mcp --help
nymrel-mcp --stdioFor an MCP client, use the absolute path to the virtual environment's nymrel-mcp executable. On Windows, that executable is under .venv\Scripts; on macOS and Linux, it is under .venv/bin.
Protocol contract
The stdio engines support two MCP behavior families without changing the exposed tool, resource, or prompt catalog.
Modern requests
Protocol revision:
2026-07-28.Discovery method:
server/discover.Each modern request carries protocol and client capability metadata in
params._meta.Successful results include
resultType: "complete"plus server identity metadata.Discovery and static list results may include bounded public cache hints.
Initialize-era compatibility
Supported initialize-era revisions are:
2024-10-072024-11-052025-03-262025-06-182025-11-25
Legacy responses remain free of modern-only fields. A modern revision offered through initialize receives the preferred legacy counteroffer rather than a false modern handshake.
Fail-closed behavior
Unsupported modern revisions return MCP error
-32022with the supported revision.Missing or malformed modern metadata returns
-32602.JSON-RPC notifications remain silent.
pingandnotifications/initializedremain initialize-era behavior.Modern list-change subscriptions are not advertised because this server does not implement
subscriptions/listen.
Exposed tools
The hub exposes Nymrel-oriented tools through one MCP surface. Repository names below identify the source project represented by each tool; they are not registry-publication claims.
# | MCP tool | Source project | Purpose |
1 |
|
| AI commerce readiness and structured-data checks. |
2 |
|
| Destructive-command and path-safety inspection. |
3 |
|
| Lease coordination and fencing generations. |
4 |
|
| Machine-readable organization and trust metadata. |
5 |
|
| Canonical signed claim receipt generation. |
6 |
|
| Public-web scrape, map, bounded crawl, and local HTML-to-Markdown extraction. |
7 |
|
| Agent liveness and heartbeat reporting. |
8 |
|
| Deterministic quote calculation. |
9 |
|
| Local-model capability inspection and routing support. |
10 |
|
| Commerce-protocol and payment-challenge utilities. |
11 |
|
| Workspace rollback, audit, and outbound-request guardrails. |
12 |
|
| Declarative decision-card rendering. |
13 |
|
| Inter-agent message envelopes and dispatch support. |
14 |
|
| Merkle receipt and signature verification. |
15 |
|
| Provider-neutral public-web search with configured fallback across Exa, Tavily, Brave Search, and SerpAPI. |
Web search gateway
nymrel_web_search is the discovery layer of the studio web-access stack. API keys are read only from the MCP server environment (EXA_API_KEY, TAVILY_API_KEY, BRAVE_SEARCH_API_KEY, SERPAPI_API_KEY); callers cannot pass credentials as tool arguments. provider: "auto" prefers Exa/Tavily-style agent search and falls back across configured providers. mode: "serp_exact" prioritizes SerpAPI, while mode: "research" asks Tavily for advanced search when routed there. Results are normalized, domain-filtered, and never synthesized when providers fail.
Search does not fetch arbitrary result URLs. Escalate selected URLs to nymrel_crawler_mesh for extraction/crawling, and use a separately privileged browser service only when JavaScript execution, login state, forms, or other interaction is genuinely required.
Crawler Mesh agent usage
nymrel_crawler_mesh is the studio-owned replacement for the paid Firecrawl path. It keeps one stable MCP tool name and accepts action: "scrape" | "map" | "crawl".
scrape: fetch one public HTTP(S) page or convert caller-supplied HTML locally.
map: discover URLs with optional search filtering and bounded sitemap/link traversal.
crawl: recursively fetch a bounded child path by default, with explicit full-domain or subdomain expansion.
Network operations reject private/loopback/link-local targets, validate DNS and every redirect destination, respect robots.txt, cap decoded response size, and enforce MCP hard caps of 100 pages and depth 10.
Responses identify
nymrel-crawler-meshas the provider and use zero Firecrawl/provider credits.Firecrawl Cloud is not a fallback and no Firecrawl key is required.
Both the TypeScript/Node and Python stdio engines expose these crawler actions. Their required upstream modules are pinned to nymrel/nymrel-crawler-mesh@02e96d9c5e32d002160f9af70f6297010a11aa73. See vendored source and license provenance.
Resources
nymrel://status— server identity, engine status, tool count, and supported protocol revisions.nymrel://ecosystem— catalog of the Nymrel project references represented by the hub.nymrel://llms-manifest— machine-readable usage and trust guidance.
Prompt templates
audit-website-ucp— guided AI commerce readiness audit.secure-agent-command— command-safety review using Nymrel guardrail tools.init-two-seat-mission— two-seat Command Studio setup.
Proof receipts and authentication
nymrel_proof_ledger requires action, agentId, an object payload, and a
caller-supplied signingKey and explicit algorithm: "HMAC-SHA256" or
"Ed25519". Ed25519 accepts a raw 32-byte hex private key; PEM is not accepted
by these portable adapters. Use a cryptographically random HMAC secret of at
least 32 bytes. The receipt is
canonical nymrel-proof-ledger Protocol 2.0.0, using RFC 8785 JSON and RFC 6962
Merkle hashing. There is no built-in secret or identity registry. MCP clients may
log request arguments, so use a client and transport appropriate for your signing
material. Prefer Ed25519 public keys for verification-only clients.
Call nymrel_proof_verify with receipt. To authenticate, supply both
publicKeyOrSecret and expectedAlgorithm (HMAC-SHA256 or Ed25519) from
trusted key configuration. Never derive the expected algorithm from a receipt;
this prevents treating a known public key as an HMAC secret.
Result | Meaning |
| No key supplied: envelope and Merkle consistency only. Signature, identity, metadata/payload and previous-hash reference are not authenticated. |
| The signature matches the supplied HMAC secret or Ed25519 public key. The caller must obtain that key independently from a trusted source. |
| Malformed/unsupported input, Merkle mismatch, or signature failure with the supplied key. |
The compatibility field verified aliases trusted. Verdicts are
INTEGRITY_ONLY, SIGNATURE_AUTHENTICATED, or INVALID; no tool claims a
receipt is universally tamper-free. agentId becomes the task runner and signed
signer label, not a verified real-world identity. payload and optional
prevProofHash are signed metadata; chain continuity and execution are not
checked. Artifact files are never read and Git commands are never executed by
these MCP adapters. Environment fields describe the server runtime only;
ATTESTED and the canonical default exitCode: 0 are claims, not execution evidence.
Missing signatures fail. Old simulated hub receipts and Protocol v1 receipts are rejected; v1 does not authenticate all v2 identity and metadata fields. Callers must migrate to canonical v2 receipts and supply their own signing key. Unknown tool arguments fail instead of being silently ignored. Extra envelope fields follow the upstream forward-compatibility contract and may be outside the signature; do not use unrecognized fields as authenticated claims.
The canonical core is vendored verbatim at a pinned commit, because it is not
available as a published dependency. See source and license provenance.
Its verification API also binds the independently configured key algorithm.
Git collection is off by default upstream and explicitly disabled by these adapters.
Both language suites run the same adversarial fixtures; after building Node and
installing Python, run python tests/proof_cross_runtime.py for schema equality
and bidirectional HMAC/Ed25519 receipt verification.
Inputs are limited to 64 nested levels, 10,000 values, and 1 MiB of canonical
JSON. Numbers must be finite and within JavaScript's safe integer magnitude
(±9007199254740991); strings must contain valid Unicode scalar values. These
limits keep the two runtimes interoperable. Proofs have no freshness/replay gate.
Security boundary
The hub exposes application-level inspection, proof, lease, and credential-pattern tools. These tools do not replace operating-system isolation, network enforcement, independent authorization, credential rotation, or sandboxing. Consumers must apply least privilege and validate tool results at the actual enforcement point.
See SECURITY.md for responsible disclosure and supply-chain controls.
Validation
The repository pins npm 12.0.2 and targets Node.js 22/24 plus Python 3.11-3.14. The Node runtime pins undici@8.12.3 for Crawler Mesh DNS/IP-pinned HTTP transport; the crawler implementation itself is vendored verbatim from a reviewed Nymrel revision with an exact Git-blob manifest. Python uses cryptography>=50.0.1,<51 for Ed25519 and rfc8785==0.1.4 for canonical JSON, matching the pinned Proof Ledger implementation.
Node.js
corepack npm@12.0.2 ci
corepack npm@12.0.2 run verify
corepack npm@12.0.2 audit --audit-level=high
corepack npm@12.0.2 audit --omit=dev --audit-level=high
corepack npm@12.0.2 install-scripts lsThe verification contract covers toolchain pins, TypeScript compilation, protocol tests, CI-policy tests, and the npm package allowlist.
Python
python -m pip install --requirement requirements-dev.txt
python -m pip install --editable .
python -m pip check
python -m ruff check python tests
python scripts/check-bandit.py
python -m pytest -q
python -m pip uninstall --yes nymrel-mcp-hub
python -m pip_audit --strict
python -m build --outdir dist-py
python -m twine check dist-py/*The Bandit gate scans all runtime sources and recognizes only the exact reviewed vendor findings recorded for the pinned Proof Ledger and Crawler Mesh sources. Changed or additional findings fail; see docs/proof-ledger/bandit-reviewed.json and docs/crawler-mesh/bandit-reviewed.json.
Local validation proves a source candidate; it does not prove a hosted check or registry publication.
Contributing
See CONTRIBUTING.md. Protocol changes must preserve TypeScript/Python parity and prove both modern and initialize-era behavior. Release or publication work requires a separate operator decision.
License and organization
Brand: Nymrel
Website: https://nymrel.com
Contact: contact@nymrel.com
Legal entity: JalenBuilds LLC
License: MIT — see LICENSE
Available Tools
14 toolsnymrel_a2ui_renderA
Generates Google A2UI v0.8 declarative JSON decision cards, diff inspectors, and parameter tables in signature Nymrel Warm Paper aesthetics (#FAF8F2, #2A332E, #A8541F).
Inputs: title (required), summary (required), fields, actions, +1 more.
nymrel_beacon_pingB
Registers agent liveness heartbeat, queries active fleet presence, and monitors multi-agent runtime health to prevent stalled/zombie processes.
Inputs: agentId (required), role, action, status, +1 more.
nymrel_crawler_meshA
High-throughput clean web crawler & Markdown AST extractor. Strips boilerplate, ads, scripts, and navigation to produce token-optimized Markdown for LLMs.
Inputs: url, html, extractMetadata.
nymrel_headless_quoteA
Calculates instant dynamic service quotes, price range estimators, and lead capture payloads with Nymrel Warm Paper presets.
Inputs: scope, preset, currency, rushDelivery, +1 more.
nymrel_local_forgeB
Probes local GPU / Ollama status, routes tasks across 3 tiers (Luna Local -> Terra Balanced -> Sol Frontier), and tracks cumulative cloud API dollar savings.
Inputs: action, taskDescription, tokensProcessed.
nymrel_machine_trustB
Generates Dual-Audience machine trust artifacts: Hierarchical Schema.org JSON-LD entity graphs (Nymrel -> JalenBuilds LLC), semantic /llms.txt manifests, and AI crawler robots.txt policies.
Inputs: domain, entityName, description, targetFormat.
nymrel_open_ucpB
Executes Universal Commerce Protocol (UCP) actions: x402 HTTP micropayment requests, AP2 multi-party price negotiation, and cart commitments.
Inputs: cart, action, maxBudgetUsd, merchantEndpoint.
nymrel_proof_ledgerC
Generates RFC-6962 compliant SHA-256 Merkle tree execution attestations, audit receipts, and verification proof paths for autonomous agent operations.
Inputs: action (required), agentId (required), payload (required), prevProofHash.
nymrel_proof_verifyA
Cryptographically verifies RFC-6962 Merkle tree execution receipts, checks proof paths, and validates digital signatures to guarantee zero tampering.
Inputs: receipt (required).
nymrel_sandstormA
Enforces Zero-Trust agent sandboxing: Copy-on-Write workspace isolation, secret exfiltration masking (API keys, JWTs), and token/financial spend rate-limiters.
Inputs: action, content, tokenSpend, maxTokenSpendLimit.
nymrel_surety_guardA
Pre-execution safety firewall intercepting destructive shell commands (rm -rf, DROP, format), path traversal violations, and unconstrained deletes before agent execution.
Inputs: command (required), strict, workingDirectory.
nymrel_swarm_busB
Dispatches structured message envelopes across the multi-agent studio task bus for inter-agent coordination (handoffs, reviews, broadcasts).
Inputs: topic (required), payload (required), toAgent (required), fromAgent (required), +1 more.
nymrel_swarm_claimA
Claims directory locks, coordinates distributed agent leases with fencing tokens, and manages two-seat Command Studio mission state.
Inputs: agentId (required), repoPath (required), role, action, +2 more.
nymrel_ucp_auditA
Audits any URL or HTML snippet for AI Agent Commerce Readiness across 7 structural layers (JSON-LD, x402 micropayments, /llms.txt, robots.txt crawler posture, AP2 negotiation).
Inputs: url, html, strictMode.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 14 tools
Each tool targets a distinct functional area—commerce, proof verification, crawling, sandboxing, swarm coordination, etc.—and the descriptions make those boundaries clear. A few pairs (proof_ledger/proof_verify, swarm_claim/swarm_bus) are semantically adjacent but not interchangeable, so the risk of selecting the wrong tool is low.
All tool names follow a consistent lowercase snake_case style with the nymrel_ prefix, which aids readability and grouping. However, the second element mixes actions ('audit', 'ping', 'render') with nouns ('ledger', 'mesh', 'bus', 'sandstorm'), so there isn't one strict verb_noun or noun_verb pattern.
At 14 tools, the server is well-scoped for a multi-purpose MCP hub that covers commerce, security, swarm coordination, proof attestation, crawling, and rendering. Each tool addresses a distinct capability without redundancy or unnecessary bloat.
The tool surface covers complete workflows within its major clusters: attestation has generation and verification, swarm has claiming, communication, and health pings, and UCP has auditing as well as execution. Minor gaps exist, such as no explicit tool for deleting or managing previously generated artifacts, but these do not create dead ends for the intended agent workflows.
Maintenance
Related MCP Connectors
Machine-native tools for autonomous agents with free routing and x402 USDC pay-per-call.
Agent-native security, trust, reliability, data and procurement tools for AI workflows.
AI agent infrastructure for discovery, authorization, execution, identity, and signed receipts.
Directory of APIs, merchants, and tools AI agents can actually use.
Related MCP Servers
- AlicenseNot gradedqualityCmaintenanceExposes 20 deterministic tools for finance, data, content, and lifestyle tasks via MCP, A2A, and REST interfaces, enabling automation and multi-agent workflows.MIT

wham-engineofficial
AlicenseCqualityCmaintenanceProvides AI agents with 200 deterministic micro-services covering code auditing, 3D/robotics, media processing, cryptography, blockchain settlement, AI safety, networking, and math, with automatic x402 USDC micropayments.200MIT
warppay402-mcpofficial
AlicenseNot gradedqualityBmaintenanceEnables AI agents to access 28 monetized Web3 developer tools, including web scraping, PDF extraction, smart contract deployment, and cross-chain USDC micropayments via SSE and x402.MIT- AlicenseNot gradedqualityBmaintenanceEnables agents and clients to manage wallet-owned identity, private per-agent context, swarm coordination, jobs and offerings, policy-gated tool execution, and audit proofs, with Robinhood Chain execution and payment verification support.MIT