Skip to main content
Glama
dressedinblack5

qwispr

by dressedinblack5
OverviewSchemaSourceRelated ServersDiscussionsAdmin
TypeScript
Local

qwispr — Hybrid Quantum/Classical Code Intelligence

Routes software engineering problems to quantum algorithms (QAOA, VQE, Grover, QWalk, QSVM) on PennyLane simulators with deterministic classical fallback.

qwispr is a CLI + MCP server that treats SE problems as optimization/quantum problems: dependency conflicts → QAOA, test generation → VQE, code search → Grover amplification, call-graph analysis → QWalk, refactoring → QWalk+QSVM. Runs locally on default.qubit/lightning.qubit — no cloud credentials required.

Table of Contents

Related MCP server: code-graph-mcp

Quickstart

# Install
npm i -g qwispr

# 1. Analyze a file (call-graph + QWalk metrics)
qwispr analyze --file src/cli.ts --entry main

# 2. Search with Grover amplification
qwispr search --pattern "eval|dangerous" --files "src/**/*.ts" --top 10

# 3. Resolve dependency conflict (QUBO → QAOA)
echo '{"Q":[[1,-2],[-2,1]]}' > qubo.json
qwispr run --task resolve --qubo qubo.json --vars 2

# 4. Generate test inputs that hit branches (VQE)
qwispr testgen --file src/utils.ts --function parseJson --layers 2

# 5. Get refactoring suggestions (QWalk+QSVM)
qwispr refactor --file src/cli.ts --top 5

# 6. Run MCP server (for VS Code / Claude / etc.)
qwispr mcp --stdio

Installation

Method

Command

npm (global)

npm i -g qwispr

npx (no install)

npx qwispr --help

Docker

docker run --rm ghcr.io/qwispr/qwispr:latest --help

From source

git clone https://github.com/qwispr/qwispr && cd qwispr && npm ci && npm run build

Requirements:

  • Node.js ≥ 20

  • Python ≥ 3.11 + PennyLane (auto-installed via pip on first quantum run)

  • Fallback: QWISPR_DEVICE=default.qubit runs pure Python (no C++ deps)

Commands

Command

Alias

Description

analyze

qwalk

Call-graph + QWalk metrics (reachability, centrality, diameter, hotSpots; diameter -1 = disconnected)

search

grover

Grover-ranked regex search over glob

vqe

—

VQE ground-state for QUBO

testgen

—

VQE boundary test inputs for a function

refactor

—

QWalk+QML refactoring candidates

run

orchestrate

Hybrid router (classical vs quantum)

hardware

backend

List backends + current device

mcp

—

stdio JSON-RPC server (5 tools)

Global options:

--backend simulator|lightning   (also QWISPR_BACKEND env)
--help, -h                      show help

Problem → Algorithm Map

SE Problem

Quantum Algorithm

Skill

How it Works

Dependency resolution (lockfile conflicts)

QAOA

qaoa-agent

QUBO → Ising Hamiltonian → RZZ/RZ + RX mixer, GradientDescent optimizer

Code search (AST/regex patterns)

Grover

grover-agent + search-agent

Analytic Grover iteration (π/4√(N/M)) amplifies matching hits

Test generation (branch coverage)

VQE

vqe-agent + testgen-agent

Branch-distance QUBO → hardware-efficient ansatz RY/RZ+CNOT → parameter-shift

Call-graph analysis (reachability, centrality)

QWalk

qwalk-agent + analyze-agent

BFS + BTree + Floyd (classical fallback; quantum coin+shift for n≤8)

Refactoring suggestions (god functions)

QWalk + QSVM

refactor-agent

Centrality × (1 − cohesion) via simulated RBF kernel

Fragment classification (buggy/clean)

QSVM

qml-agent

Token/AST/n-gram → RY angles → RBF `

MCP Server

qwispr exposes 5 tools via stdio JSON-RPC (MCP protocol):

# Start server
qwispr mcp --stdio

# List tools
echo '{"jsonrpc":"2.0","id":1,"method":"tools/list"}' | qwispr mcp --stdio

# Call analyze
echo '{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"analyze","arguments":{"file":"src/cli.ts"}}}' | qwispr mcp --stdio

Tool

Description

analyze

Call-graph + QWalk metrics

search

Grover-ranked regex search

testgen

VQE boundary test generation

refactor

QWalk+QML refactoring candidates

hardware

List backends + current device

Register with OpenAxe:

openaxe mcp add qwispr --cwd . -- node dist/cli.js mcp --stdio
openaxe mcp list  # shows qwispr ✓ connected

VS Code Extension

cd extensions/vscode
npm ci && npm run build    # produces dist/extension.js
vsce package               # creates qwispr-vscode-0.1.0.vsix
code --install-extension qwispr-vscode-0.1.0.vsix

Commands:

  • qwispr.analyze — call-graph + QWalk webview (hotSpots table)

  • qwispr.search — quickpick Grover-ranked results

  • qwispr.testgen — JSON input → boundary test cases

Benchmarks

# Synthetic QUBOs (n=3..8, LCG-generated)
BENCH_N=5 npm run benchmark

# Real lockfile conflicts (react/webpack/eslint/typescript)
BENCH_N=5 npm run benchmark:real

# Generate comparison report
npm run benchmark:report  # → report/index.md + report/data.json

Results (BENCH_N=2, deterministic seed):

Suite

Cases

Success Rate

Avg Time

p95

synthetic

2

100%

~4s

~4.5s

real

2

100%

~4s

~4.5s

Speedup: lightning.qubit ~2-3× faster than default.qubit (benchmarks/BENCH_DEVICE_LIGHTNING.md).

QWISPR_BACKEND=lightning BENCH_N=5 npm run benchmark

Environment Variables

Variable

Default

Description

QWISPR_BACKEND

simulator

simulator | lightning

QWISPR_DEVICE

—

Direct PennyLane device string (e.g., lightning.qubit)

QWISPR_QPU_SHOTS

1024

Shots for QPU (clamp 1..10000)

QWISPR_LAYERS

2

VQE/QAOA ansatz layers (honored by python workers/benchmark only; CLI --layers overrides)

QWISPR_ITERS

50

Optimizer iterations (honored by python workers/benchmark only; CLI --iters overrides)

QWISPR_TELEMETRY

0

Set 1 to enable ~/.qwispr/telemetry.jsonl

QWISPR_TELEMETRY_PATH

~/.qwispr/telemetry.jsonl

Custom telemetry path

QWISPR_CALIBRATION

1.0

Reserved/unused (no runtime effect; no reader in code)

QWISPR_QPU_DRYRUN

0

Set 1 for dry-run QPU path

QWISPR_ALLOW_ABSOLUTE

0

Bypass workspace-root jail (search globs + analyze/refactor/testgen --file)

QISKIT_TOKEN

—

IBM Quantum token, reported in hardware status (hasToken); live QPU calls not implemented — set QWISPR_QPU_DRYRUN=1 for the simulated path

Architecture

┌─────────────────────────────────────────────────────────────┐
│                    SOURCE / LOCKFILE / QUBO                 │
└─────────────────────────────────────────────────────────────┘
                            │
        ┌───────────────────┼───────────────────┐
        ▼                   ▼                   ▼
┌───────────────┐   ┌───────────────┐   ┌───────────────┐
│  code-graph   │   │problem-encoder│   │  (direct QUBO)│
│  (regex AST)  │   │(branch-dist)  │   │               │
└───────┬───────┘   └───────┬───────┘   └───────┬───────┘
        │                   │                   │
        └───────────────────┼───────────────────┘
                            ▼
              ┌───────────────────────────┐
              │      ORCHESTRATOR         │
              │  nVars ≤ threshold?       │
              │  classical : quantum      │
              └───────────┬───────────────┘
                          │
          ┌───────────────┴───────────────┐
          ▼                               ▼
   ┌─────────────┐                 ┌─────────────┐
   │  CLASSICAL  │                 │  QUANTUM    │
   │  (brute n≤4)│                 │  PennyLane  │
   └──────┬──────┘                 │  • QAOA     │
          │                        │  • VQE      │
          │                        │  • Grover   │
          │                        │  • QWalk    │
          │                        │  • QSVM     │
          └───────────────┬────────┘
                          ▼
              ┌───────────────────────────┐
              │     BITSTRING + ENERGY    │
              └───────────┬───────────────┘
                          ▼
              ┌───────────────────────────┐
              │        DECODER            │
              │  patch / inputs / hits    │
              └───────────────────────────┘

Hybrid routing threshold adapts via telemetry (QWISPR_TELEMETRY=1):

  • Quantum 2× slower with no gain → threshold++

  • Quantum success rate > classical + 0.1 → threshold--

  • Clamped to [2, 8]

Ponytail Minimalism

This project follows ponytail full — minimal, stdlib-first, no unnecessary abstractions.

Shortcut

Ceiling

Upgrade Path

Manual JSON-RPC

no SDK

@modelcontextprotocol/sdk if client fails

Regex call-graph

no tree-sitter

native parser when lands

BFS ≈ QWalk

classical fallback

true quantum walk n≤8

Heuristic ReDoS guard

nested quantifier only

re2 if exposed over network

Adaptive threshold

heuristic 2×/rate

ML when >10k events/day

Every // ponytail: comment marks a deliberate ceiling with upgrade path.

Stats:

  • 0 runtime dependencies (dependencies: {})

  • ~250 lines removed via audit

  • 7 dev deps removed (chalk, ora, fast-json-patch, tree-sitter*, semver, yaml)

  • 16/16 tests pass, 0 vulnerabilities

Contributing

# 1. Fork & clone
git clone https://github.com/yourfork/qwispr
cd qwispr

# 2. Install deps
npm ci

# 3. Run checks
npm run lint      # eslint
npm run typecheck # tsc --noEmit
npm test          # vitest 16/16

# 4. Benchmarks (optional)
npm run benchmark
npm run benchmark:real

# 5. Commit (conventional commits)
git commit -m "feat: add new quantum algorithm"

# 6. PR → CI runs lint + typecheck + test + benchmark:smoke

Code style: Strict TypeScript, no any, no ts-ignore, minimal deps, ponytail ceiling comments.

License

MIT — see LICENSE for details.

This server cannot be deployed

One endpoint for every MCP toolRayrun runs every MCP server behind one URL and holds the credentials. Each call is decided per client, tool, and argument, then recorded.ray.runAd

Maintenance

ActivityMaintained
ResponsivenessNo issues

Resources

Related MCP Connectors

Related MCP Servers