One endpoint, hundreds of tools
Agents connect once over stdio or Streamable HTTP. The gateway serves the full nexus.* surface — register, remove, inspect, list, route, discover, invoke, approvals.
Connect once. Discover dynamically. Route intelligently. Execute safely.
One umbrella MCP endpoint in front of hundreds of tools — the agent only ever sees the right capability at the right time.
Agents connect once over stdio or Streamable HTTP. The gateway serves the full nexus.* surface — register, remove, inspect, list, route, discover, invoke, approvals.
Instead of dumping 500 tool schemas into the agent's context, nexus.discover returns the minimal tool surface that fits the current request.
Every decision carries a provider, confidence, matched capabilities, and alternatives — so you can see why a tool was chosen.
Allow / deny / approval gates with per-tool permission scopes. Approval-gated tools queue for an operator and resolve through the gateway or the dashboard.
Heuristic → semantic → LLM fallback chain. The zero-dependency stack routes perfectly offline; Gemini and OpenRouter bolt on only when you add a key.
A 32-task §31 reference suite runs offline in CI. mcp-nexus benchmark exits 0 only when every exact + semantic task routes correctly.
AI AGENT
Claude / Cursor / Codex
│
▼
┌─────────────────┐
│ MCP SERVER │ stdio · Streamable HTTP
└────────┬────────┘
│
▼
┌─────────────────┐
│ DISCOVERY │
└────────┬────────┘
│
┌─────────┴─────────┐
▼ ▼
TOOL REGISTRY ROUTER intent overlay (§25)
│
┌───────────┼───────────┐
▼ ▼ ▼
Heuristic Semantic LLM*
│ │ │
└───────────┼───────────┘
▼
POLICY ENGINE
│
▼
TOOL RUNNER
│
local · stdio · docker · http
│
▼
MCP TOOL
* LLM is optional — MCP Nexus routes perfectly with no API key, no GPU, no network.
Run everything with npm install && npm start and a reference dashboard over mcp-nexus dashboard.
6 tools · 32 tasks (exact / semantic / ambiguous / unknown) · fully offline · asserted in CI
| Provider | Accuracy | exact | semantic | ambiguous | unknown |
|---|---|---|---|---|---|
| Heuristic | 100% | 15/15 | 4/4 | 5/5 | 8/8 |
| Semantic | 87.5% | 15/15 | 4/4 | 2/5 | 7/8 |
| Hybrid chain | 93.8% | 15/15 | 4/4 | 4/5 | 7/8 |
| Large (50 tools) | 100% | 20/20 | — | — | — |
Hard failures: none. The benchmark command is the CI gate.
$ git clone https://github.com/dsk-dev-ai/mcp-nexus.git
$ cd mcp-nexus
$ npm install
# stdio for a local client, or Streamable HTTP for remote clients:
$ npm start # stdio
$ npm start -- start:http # http://127.0.0.1:3001/mcp
# manage tools, route, and verify in a second terminal:
$ npm start -- add tools/repoarch.json
$ npm start -- search "check vulnerable dependencies"
$ npm start -- doctor
$ npm start -- benchmark # exit 0 = §31 reference green
MCP gateway (stdio + HTTP), dashboard REST, CLI, SDK.
System design, router chain, policy, executor.
Heuristic, semantic, LLM providers, intent overlay.
Manifest schema (JSON + YAML), transports, permissions.
Full §31 numbers, method, reproduction.
Web UI + REST, policies, approvals, auth.
Threat model, permissions, audit log.
Verified MCP client compatibility matrix.
Build Nexus-compatible tools and routers.
V1 latency targets vs measured.
Docker, CI, environment setup.
Every release, documented.
Sponsorships fund LLM-provider test keys, CI minutes, and documentation.
Sponsor the project →