runtime-mcp-connect
Exposes a live application's runtime state to coding agents, including logs, DB schema, config, and feature flags, with read-only SQL and secret redaction.
README
<div align="center">
runtime-mcp
Your app's live runtime state — logs, DB schema, config, feature flags — as MCP tools for coding agents.
Drop it into your app. Your agent stops guessing.
</div>
When opencode, Claude Code, or Cursor debug your app today, they work blind: stale guesses about your schema, pasted log fragments, invented config values.
runtime-mcp closes that gap. Two lines of setup and any MCP-speaking agent can query what your running process actually sees:
Agent: "why is checkout failing?"
├─ get_errors → TypeError at checkout.ts:42, full stack trace
├─ db_query → orders table has no `coupon_id` column
├─ config:// → STRIPE_KEY: [REDACTED], TAX_RATE: 0 ← there it is
└─ "Found it — you renamed the env var."
Quickstart
0 — Install:
npm install runtime-mcp runtime-mcp-integrations
1 — Instrument your app (one import):
import { runtimeMcp } from 'runtime-mcp';
import { pgAdapter } from 'runtime-mcp-integrations/pg';
await runtimeMcp({
db: pgAdapter(pool),
config: process.env,
flags: myFlagRegistry,
});
2 — Register the shim in your project's .mcp.json:
{
"mcpServers": {
"runtime": { "command": "npx", "args": ["-y", "runtime-mcp-connect"] }
}
}
3 — Start your app. That's it. The agent connects automatically and sees live state from then on.
What agents get
Tools
| Tool | Description |
|---|---|
get_logs |
Recent logs · filter by level, free-text query, time window |
get_errors |
Error entries with stack traces — first call when something broke |
db_query |
Read-only SQL (SELECT/WITH only) · writes rejected · auto LIMIT |
app_info |
Name, version, PID, uptime, registered routes |
Resources
| Resource | Description |
|---|---|
config://current |
Live config with secrets replaced by [REDACTED] |
flags://state |
Feature flag states per environment |
schema://tables |
Table inventory |
schema://table/{name} |
Columns, primary keys, indexes, foreign keys |
Architecture
┌──────────────────────────────┐ ┌─────────────────────────────┐
│ Your App │ │ Agent (opencode/Claude) │
│ │ │ │
│ runtimeMcp({ db, config }) │ │ .mcp.json │
│ ├─ log ring buffer │ │ └─ runtime-mcp-connect │
│ ├─ schema introspection │ │ │ stdio │
│ ├─ config + redaction │◄─────────┼──────────┘ │
│ └─ feature flags │ loopback│ │
│ │ HTTP │ │
│ 127.0.0.1:<port>/mcp │ │ │
│ .runtime-mcp.json ──────────┼─ pid·url·token·(chmod 600) │
└──────────────────────────────┘ └─────────────────────────────┘
Why two pieces? Agents spawn their own processes — they can't reach into your app. The discovery file bridges them: the library writes it on boot, the shim reads it on every call. App restarted with a new port or token? The shim reconnects automatically. Nothing to reconfigure, ever.
Adapters
npm install runtime-mcp runtime-mcp-integrations
<table> <tr><th>Import</th><th>Use with</th></tr> <tr><td><code>runtime-mcp-integrations/pg</code></td><td><code>pg</code> Pool — full introspection + queries</td></tr> <tr><td><code>runtime-mcp-integrations/prisma</code></td><td>PrismaClient (Postgres)</td></tr> <tr><td><code>runtime-mcp-integrations/node-sqlite</code></td><td><code>node:sqlite</code> DatabaseSync — zero native deps</td></tr> <tr><td><code>runtime-mcp-integrations/pino</code></td><td>Pino destination mirroring into the buffer</td></tr> <tr><td><code>runtime-mcp-integrations/winston</code></td><td>Winston transport</td></tr> <tr><td><code>runtime-mcp-integrations/express</code></td><td>Error middleware recording request failures</td></tr> </table>
Structured logging example:
import pino from 'pino';
import { createRuntimeMcpPinoDestination } from 'runtime-mcp-integrations/pino';
const logger = pino(pino.multistream([
pino.destination(1),
createRuntimeMcpPinoDestination(),
]));
No logger wired up? Console capture is on by default — console.error and friends flow into the same buffer.
Security model
Built for local development and staging. Defense in depth by default:
| Control | Behavior |
|---|---|
| Loopback bind | Endpoint listens on 127.0.0.1 only; remote exposure requires explicit opt-in |
| Token auth | Bearer token per app start, stored chmod 600 in .runtime-mcp.json |
| Read-only SQL | Allowlist parser: non-SELECT rejected, stacked statements rejected, keywords inside strings/comments ignored |
| Secret redaction | Password/token/key/auth-shaped keys and known token formats become [REDACTED] before leaving the process |
| No telemetry | Zero phone-home |
Add .runtime-mcp.json to .gitignore.
Why now
| Existing solution | Limitation |
|---|---|
Next.js 16 /_next/mcp |
Framework-locked |
| Vercel runtime logs MCP | Cloud platform only |
| Datadog / Sentry / Supabase MCPs | SaaS accounts; not your live dev process |
Every platform is building this inward. Nobody built it generically. This is that.
Demo
packages/demo-app is an Express + SQLite store with two planted bugs — an N+1 query on /orders, and a misnamed DEMO_TAX_RATE env var that silently zeroes checkout tax. Point your agent at it and ask "why are checkout totals wrong?"
pnpm install && pnpm -r build
pnpm --filter @runtime-mcp/demo-app dev
Packages
All packages are published to npm (v0.1.0):
| Package | npm | Purpose |
|---|---|---|
runtime-mcp |
npm ↗ | In-process registry · loopback endpoint · discovery · redaction |
runtime-mcp-connect |
npm ↗ | stdio shim agents register in .mcp.json |
runtime-mcp-integrations |
npm ↗ | Framework, logger, and database adapters |
@runtime-mcp/demo-app |
— | Debugging demo with planted bugs (repo only) |
Requires Node ≥ 20. ESM.
Development
pnpm install
pnpm -r build
pnpm test # unit + end-to-end over real HTTP and stdio transports
<details> <summary><strong>Roadmap</strong></summary>
- [ ] Live log tailing via MCP resource subscriptions
- [ ] Fastify / Hono adapters
- [ ] MySQL / Drizzle introspection adapters
- [ ] Python SDK parity (FastAPI / Django)
- [ ] Multi-project workspaces in the connect shim
</details>
License
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.