☕ MCPresso

Brew your MCP server in under 60 seconds.

MCPresso is a production-grade Python package that transforms natural language descriptions into fully validated, security-audited MCP (Model Context Protocol) servers. It combines LLM-powered code generation with a 5-category static analysis engine, iterative auto-repair, composite confidence scoring, semantic template reuse, and automatic test suite co-generation — all in a single brew command.

---

Architecture

╔══════════════════════════════════════════════════════════════════════╗
║                        MCPresso Pipeline                            ║
║                                                                      ║
║  Natural Language Description                                        ║
║         │                                                            ║
║         ▼                                                            ║
║  ┌─────────────┐    ┌───────────────────────────────────────────┐   ║
║  │  Registry   │───▶│           generator.py                    │   ║
║  │  Search     │    │  (ADAPT ~10s / SEED ~30s / FULL ~60s)     │   ║
║  │  (semantic) │    └───────────────┬───────────────────────────┘   ║
║  └─────────────┘                    │                                ║
║                                     ▼                                ║
║                          ┌─────────────────────┐                    ║
║                          │   validator.py       │                    ║
║                          │  5 Categories:       │                    ║
║                          │  1. Structural       │                    ║
║                          │  2. Protocol         │                    ║
║                          │  3. Security         │                    ║
║                          │  4. Robustness       │                    ║
║                          │  5. Documentation    │                    ║
║                          └──────────┬──────────┘                    ║
║                                     │                                ║
║                                     ▼                                ║
║                          ┌─────────────────────┐                    ║
║                          │    repair.py         │◀─── (≤3 iters)    ║
║                          │  Auto-fix criticals  │                    ║
║                          └──────────┬──────────┘                    ║
║                                     │                                ║
║                                     ▼                                ║
║                          ┌─────────────────────┐                    ║
║                          │    scorer.py         │                    ║
║                          │  Composite Score:    │                    ║
║                          │  • Validation  40%   │                    ║
║                          │  • Complexity  20%   │                    ║
║                          │  • Consistency 20%   │                    ║
║                          │  • Security    20%   │                    ║
║                          └──────────┬──────────┘                    ║
║                                     │                                ║
║                          ┌──────────▼──────────┐                    ║
║                          │    testgen.py        │  (optional)        ║
║                          │  3 tests/tool:       │                    ║
║                          │  happy/edge/security │                    ║
║                          └──────────┬──────────┘                    ║
║                                     │                                ║
║                          ┌──────────▼──────────┐                    ║
║                          │    registry.py       │                    ║
║                          │  Save if score ≥ 75  │                    ║
║                          └─────────────────────┘                    ║
║                                                                      ║
║         Output: server.py  +  test_server.py  +  BrewResult         ║
╚══════════════════════════════════════════════════════════════════════╝

---

Quickstart

Installation

pip install mcpresso

Or from source:

git clone https://github.com/your-org/mcpresso
cd mcpresso
pip install -e ".[dev]"

Configure API Key

cp .env.example .env
# Edit .env and set ANTHROPIC_API_KEY=sk-ant-...

Brew Your First MCP Server

# Generate a GitHub issue tracker server in under 60 seconds
mcpresso brew "A server that fetches GitHub issues and summarizes them" \
  --output github_server.py \
  --repair

# Generate with test suite
mcpresso brew "A PostgreSQL query server with connection pooling" \
  --output pg_server.py \
  --with-tests

# Validate an existing server
mcpresso validate pg_server.py

# Validate with JSON output for CI pipelines
mcpresso validate pg_server.py --report json > report.json

# Auto-repair a broken server
mcpresso repair broken_server.py --output fixed_server.py

# Generate tests for an existing server
mcpresso testgen my_server.py --output tests/test_my_server.py

# Browse the registry
mcpresso registry list
mcpresso registry search "slack notification"
mcpresso registry export backup.json

---

CLI Commands Reference

brew

Generate a production-ready MCP server from a plain English description.

# Basic brew (auto-repair enabled by default)
mcpresso brew "A server that adds, subtracts, multiplies and divides numbers"

# Save to a specific file
mcpresso brew "A server that fetches GitHub issues" --output github_server.py

# Generate server + companion client for immediate testing
mcpresso brew "A todo list server with create, list, complete, and delete tools" \
    --output todo_server.py --with-client

# Generate server + test suite
mcpresso brew "A weather forecast server" --output weather_server.py --with-tests

# Generate everything: server + client + tests
mcpresso brew "A URL shortener server" --output url_server.py --with-client --with-tests

# Skip auto-repair (faster, less polished)
mcpresso brew "A simple echo server" --no-repair

# Verbose output with per-category validation scores
mcpresso brew "A Slack notification server" --verbose

# Pass API key directly (overrides .env)
mcpresso brew "A database query server" --api-key sk-ant-xxxx

repair

Auto-repair critical issues in an existing server file (up to 3 LLM iterations).

# Repair in-place (overwrites the file)
mcpresso repair broken_server.py

# Repair to a new file
mcpresso repair broken_server.py --output fixed_server.py

# Verbose: show each fix applied
mcpresso repair broken_server.py --verbose

testgen

Generate a pytest test suite for an existing MCP server.

# Auto-detect output path (creates test_<filename>.py)
mcpresso testgen calc_server.py
# Output: test_calc_server.py

# Specify output path
mcpresso testgen todo_server.py --output tests/test_todo.py

# Generate and run the tests
mcpresso testgen calc_server.py
pytest test_calc_server.py -v

registry

Manage the local server template registry (~/.mcpresso/registry/).

# List all previously brewed servers
mcpresso registry list

# Show more entries
mcpresso registry list --limit 50

# Semantic similarity search
mcpresso registry search "github issues"
mcpresso registry search "database query"
mcpresso registry search "todo list management"

# Show registry statistics
mcpresso registry stats

# Export entire registry to JSON (for sharing or paper benchmarks)
mcpresso registry export registry_export.json

Python API

from mcpresso.pipeline import MCPressoPipeline

pipeline = MCPressoPipeline()

result = pipeline.brew(
    description="A server that queries PostgreSQL and returns query results",
    auto_repair=True,
    output_path="./pg_server.py",
    with_tests=True,
)

print(f"✅ Score: {result.final_score:.1f}/100")
print(f"📊 Tier: {result.readiness_tier.value}")
print(f"⏱  Time: {result.total_time_ms/1000:.1f}s")
print(f"🔧 Repairs: {result.repair_result.repair_iterations if result.repair_result else 0}")
print(f"🧪 Tests: {result.test_result.test_count if result.test_result else 0}")

---

CLI Reference

mcpresso brew

mcpresso brew "description" [OPTIONS]

Options:
  --output, -o TEXT      Output file path (e.g., server.py)
  --repair/--no-repair   Auto-repair critical issues (default: enabled)
  --with-tests           Generate pytest suite alongside the server
  --verbose, -v          Show detailed category breakdown
  --api-key TEXT         Anthropic API key (overrides env var)

Brewing sequence output:

  ⚙ Grinding beans... (generating from scratch ~60s)
  🔍 Checking the brew... (validating)
  🔧 Fixing the blend... (auto-repairing)
  📊 Measuring the roast... (scoring)
  🧪 Running quality checks... (generating tests)
  ✅ Your server is ready!

  ╭─ ☕ Brew Complete — a1b2c3d4 ────────────────────────────────╮
  │  Score:          ████████████████░░░░ 82.4/100              │
  │  Readiness:      ✅ STAGING_READY                           │
  │  Brew Time:      ✓ 43.2s  (under 60s ✓)                    │
  │  Tools Generated: 3                                          │
  │  Registry Match:  FULL_GENERATION                           │
  │  Repair:         1 iteration(s) | 61.3 → 82.4              │
  │  Tests:          9 tests | 3 security | ~71% coverage       │
  │  Output:         ./pg_server.py                             │
  ╰──────────────────────────────────────────────────────────────╯

mcpresso validate / mcpresso taste

mcpresso validate server.py [--report rich|json]
mcpresso taste server.py        # alias for validate

mcpresso repair

mcpresso repair broken.py [--output fixed.py]

mcpresso testgen

mcpresso testgen server.py [--output tests/test_server.py]

mcpresso registry

mcpresso registry list [--limit N]
mcpresso registry search "query"
mcpresso registry export output.json
mcpresso registry stats

Sample Brew Queries

Copy any of these directly into mcpresso brew "...":

Simple / Quick (~20–40s)

mcpresso brew "A calculator server with add, subtract, multiply, and divide" \
    --output calc_server.py --with-client

mcpresso brew "A server that converts temperatures between Celsius and Fahrenheit" \
    --output temp_server.py --with-client

mcpresso brew "A server that returns the current UTC time and converts Unix timestamps" \
    --output time_server.py --with-client

mcpresso brew "A text utilities server with word count, character count, and reverse string" \
    --output text_server.py --with-client

Medium Complexity (~40–60s)

mcpresso brew "A todo list server with create, list, complete, delete, and update tools" \
    --output todo_server.py --with-client

mcpresso brew "A file system server that reads, writes, lists, and deletes files safely" \
    --output fs_server.py --with-client

mcpresso brew "A JSON validator and formatter server with schema validation support" \
    --output json_server.py --with-client

mcpresso brew "A URL utilities server that validates URLs, extracts domains, and checks redirects" \
    --output url_server.py --with-client

Advanced / Multi-Tool (~50–90s)

mcpresso brew "A GitHub integration server that lists repos, fetches issues, and creates comments using the GitHub REST API" \
    --output github_server.py --with-client --with-tests

mcpresso brew "A PostgreSQL database server that executes queries, lists tables, and describes schemas with connection pooling" \
    --output postgres_server.py --with-client

mcpresso brew "A Slack notification server that sends messages, creates channels, and lists workspace members" \
    --output slack_server.py --with-client --with-tests

mcpresso brew "A weather server that gets current conditions and 5-day forecasts using the OpenWeatherMap API" \
    --output weather_server.py --with-client

mcpresso brew "An email server with send, read inbox, search, and draft capabilities using SMTP and IMAP" \
    --output email_server.py --with-client

---

Module Reference

Module	Purpose
mcpresso/models.py	All dataclasses and enums (single source of truth)
mcpresso/generator.py	NL → MCP server code via Claude API
mcpresso/validator.py	5-category static analysis engine (no API calls)
mcpresso/repair.py	Iterative auto-repair via Claude API (max 3 iterations)
mcpresso/scorer.py	Composite 4-component confidence scoring
mcpresso/registry.py	Persistent semantic template registry
mcpresso/testgen.py	Automatic pytest suite co-generation
mcpresso/pipeline.py	End-to-end orchestration pipeline
mcpresso/cli.py	Typer + Rich terminal interface
mcpresso/benchmark.py	20-case empirical benchmark harness

---

Validation Categories

#	Category	Weight	What it checks
1	Structural Integrity	20%	Valid Python syntax, MCP SDK imports, Server() instantiation, tool handlers registered, if __name__ == "__main__" present
2	Protocol Compliance	25%	Tools have name/description/inputSchema, return types match MCP spec, async/await patterns, McpError with correct codes
3	Security Posture	25%	No hardcoded secrets/API keys, no eval()/exec(), env vars for credentials, input validation, no path traversal
4	Robustness	20%	try/except around external calls, no bare except:, timeout handling, resource cleanup
5	Documentation	10%	Docstrings on handlers, human-readable tool descriptions, logging statements, type hints

---

Readiness Tiers

Tier	Score	Critical Issues	Description
🏆 PRODUCTION_READY	≥ 90	0	Deploy to production
✅ STAGING_READY	75–89	0	Safe for staging/testing
⚠️ DEVELOPMENT_ONLY	50–74	any	Development use only
❌ NEEDS_REPAIR	< 50	any	Requires repair before use

---

Registry Reuse Tiers

The semantic registry enables "agentic memory" — MCPresso learns from its own outputs:

Similarity	Mode	Est. Time	How it works
≥ 0.85	ADAPT	~10s	Modifies only differing parts of existing server
0.60–0.85	SEED	~30s	Uses similar server as few-shot example
< 0.60	FULL_GENERATION	~60s	Generates from scratch

---

Benchmark Results (Placeholder)

Run the benchmark suite to populate this table:

python -c "
from mcpresso.benchmark import MCPressoBenchmark, BENCHMARK_CASES
bench = MCPressoBenchmark(output_dir='./benchmark_results')
report = bench.run(cases=BENCHMARK_CASES)
bench.save_report(report)
bench.print_summary(report)
"

Metric	Value
Success Rate (STAGING_READY+)	—
P50 Generation Latency	—
P95 Generation Latency	—
Mean Validation Score	—
Repair Convergence Rate	—
Token Efficiency (tokens/success)	—
Security Detection Rate	—
Registry Reuse Rate	—
Mean Test Coverage (est.)	—

Run mcpresso benchmark with your API key to populate.

---

Environment Variables

Variable	Default	Description
ANTHROPIC_API_KEY	(required)	Anthropic API key
MCPRESSO_MODEL	claude-sonnet-4-20250514	Model to use
MCPRESSO_REGISTRY_DIR	~/.mcpresso/registry	Registry directory
MCPRESSO_SIMILARITY_THRESHOLD_ADAPT	0.85	ADAPT mode threshold
MCPRESSO_SIMILARITY_THRESHOLD_SEED	0.60	SEED mode threshold
MCPRESSO_LOG_LEVEL	WARNING	Logging level

---

Running Tests

# Install dev dependencies
pip install -e ".[dev]"

# Run all tests (no API calls — all LLM interactions are mocked)
pytest tests/ -v

# Run with coverage
pytest tests/ --cov=mcpresso --cov-report=term-missing

# Run specific test module
pytest tests/test_validator.py -v
pytest tests/test_scorer.py -v
pytest tests/test_registry.py -v

---

Design Principles (for Research)

MCPresso is designed as a research artifact for empirical study of LLM-assisted code generation quality. Key design patterns documented in the codebase:

1. Iterative Refinement Loop (pipeline.py): brew → validate → repair constitutes an agentic self-correction cycle, analogous to self-refine (Madaan et al., 2023).

2. Composite Ensemble Scoring (scorer.py): Four orthogonal quality signals (validation, complexity, consistency, security) are combined with learned weights, inspired by ensemble methods.

3. Semantic Memory & Reuse (registry.py): Dense retrieval using sentence-transformers enables few-shot grounding from prior successful generations — a novel contribution for MCP server generation.

4. Co-Generation (testgen.py): Simultaneous server + test suite synthesis using shared tool definitions, addressing the "no tests for generated code" deployment risk.

5. Separation of Concerns: Each module has a single well-defined responsibility with clean dataclass interfaces, enabling independent benchmarking of each stage.

---

License

MIT License — see LICENSE.

---

Citation

If you use MCPresso in academic work, please cite:

@software{mcpresso2025,
  author  = {Sanjoy Kumar},
  title   = {MCPresso: Brewing Production-Ready MCP Servers in Under 60 Seconds},
  year    = {2026},
  note    = {Python package. \url{https://github.com/imsanjoykb/mcpresso}},
  version = {1.1.0}
}