Mcpkit Data

The ultimate MCP server for data engineers who want to stop context-switching and start shipping ✨

A comprehensive Model Context Protocol server that gives your AI assistant superpowers to work with Kafka, databases, AWS, data pipelines, and more. Stop jumping between terminals, dashboards, and docs. Just ask your AI to do it.

Python 3.10+ License: MIT

---

⚠️ Early Days Disclaimer

Hey there! 👋 Before you dive in, here's the deal:

This is a brand new project fresh out of the oven 🍞. We're talking:

• 🆕 First release - Like, literally just born
• 🧪 Not production-ready - More like "works on my machine" ready
• 👥 Limited testing - It's mostly been tested by... well, me
• 🛠️ Actively evolving - Things might break, change, or surprise you

What this means:

• ✅ Great for experimentation and learning
• ✅ Perfect for side projects and personal use
• ✅ Awesome if you want to contribute and shape the future
• ❌ Probably not ideal for critical production systems (yet!)
• ❌ Might have bugs, quirks, or "interesting" behaviors
• ❌ Documentation might be incomplete or confusing

But here's the cool part: You can help make it better! 🚀 Found a bug? Report it! Want a feature? Ask for it! Want to contribute? We'd love that!

TL;DR: This is the "early access" version. Use at your own risk, have fun, and help us make it awesome! 🎉

---

🎯 What is this?

mcpkit-data is your AI assistant's Swiss Army knife for data engineering. It exposes 66+ tools through the MCP protocol, letting your AI:

• 🔍 Debug Kafka pipelines without leaving your editor
• 📊 Query databases (JDBC, Athena) and analyze results
• 🐼 Process data with pandas, polars, and DuckDB
• ☁️ Work with AWS (Athena, S3, Glue) seamlessly
• 🏗️ Inspect infrastructure (Nomad, Consul) in real-time
• 📈 Generate charts and evidence bundles automatically
• ✅ Validate data quality with Great Expectations

All without you writing a single line of code. Just chat with your AI. 🎉

---

⚡ Quick Start

Installation

pip install -e ".[dev]"

Run the Server

python -m mcpkit.server

Configure Cursor (or your MCP client)

Add to your Cursor MCP settings:

{
  "mcpServers": {
    "mcpkit-data": {
      "command": "python",
      "args": ["-m", "mcpkit.server"],
      "env": {
        "MCPKIT_KAFKA_BOOTSTRAP": "localhost:9092",
        "MCPKIT_SCHEMA_REGISTRY_URL": "http://localhost:8081",
        "MCPKIT_JDBC_DRIVER_CLASS": "org.postgresql.Driver",
        "MCPKIT_JDBC_URL": "jdbc:postgresql://localhost/test",
        "MCPKIT_JDBC_JARS": "/path/to/postgresql.jar",
        "MCPKIT_ROOTS": "/allowed/path1:/allowed/path2"
      }
    }
  }
}

That's it! Your AI can now use all 66+ tools. 🎊

---

🛠️ Tools Overview

📦 Dataset Registry (4 tools)

Your data's home base. Store, query, and manage datasets as Parquet files.

Tool	What it does	When to use
dataset_list	📋 List all datasets	See what data you have
dataset_info	ℹ️ Get dataset metadata	Check schema, row count
dataset_put_rows	💾 Store rows as Parquet	Save query results
dataset_delete	🗑️ Delete a dataset	Clean up old data

---

🎧 Kafka Tools (8 tools)

Debug, consume, and analyze Kafka topics like a pro.

Tool	What it does	When to use
kafka_list_topics	📝 List all topics	Discover what's available
kafka_offsets	📊 Get partition offsets	Check lag, find positions
kafka_consume_batch	📥 Consume messages → dataset	Extract data for analysis
kafka_consume_tail	🔚 Get last N messages	Debug recent events
kafka_filter	🔍 Filter with JMESPath	Find specific records
kafka_flatten	📐 Flatten nested records	Normalize JSON structures
kafka_groupby_key	🔑 Group by extracted key	Aggregate by message key
kafka_describe_topic	📋 Get topic config/partitions	Inspect topic metadata

---

📋 Schema Registry & Decoding (4 tools)

Work with Avro, Protobuf, and schema discovery.

Tool	What it does	When to use
schema_registry_get	📄 Get schema by ID/subject	Fetch schema definitions
schema_registry_list_subjects	📚 List all subjects	Discover available schemas
avro_decode	🔓 Decode Avro messages	Parse binary Avro data
protobuf_decode	🔓 Decode Protobuf messages	Parse binary Protobuf data

---

🗄️ Database Tools (2 tools)

Query databases safely with read-only enforcement.

Tool	What it does	When to use
db_query_ro	🔍 Execute read-only SQL	Query any database
db_introspect	🔎 Introspect schema	Discover tables/columns

Safety: Automatically blocks DROP, INSERT, UPDATE, DELETE, etc. ✅

---

☁️ AWS Tools (9 tools)

Query Athena, list S3, manage Glue partitions.

Tool	What it does	When to use
athena_start_query	🚀 Start SQL query	Run analytics queries
athena_poll_query	⏳ Check query status	Monitor execution
athena_get_results	📊 Get query results	Fetch data
athena_explain	📖 Explain query plan	Optimize queries
athena_partitions_list	📂 List partitions	Check table structure
athena_repair_table	🔧 Repair table (MSCK)	Fix partition metadata
athena_ctas_export	📤 CTAS export	Export to S3
s3_list_prefix	📁 List S3 objects	Browse buckets

Safety: SQL validation blocks dangerous operations by default. ✅

---

🐼 Pandas Tools (12 tools)

The data scientist's best friend, now in your AI assistant.

Tool	What it does	When to use
pandas_from_rows	📊 Create DataFrame	Build datasets
pandas_describe	📈 Get statistics	Understand your data
pandas_groupby	🔢 Group + aggregate	Summarize by categories
pandas_join	🔗 Join datasets	Combine data sources
pandas_filter_query	🔍 Filter rows	Subset your data
pandas_filter_time_range	⏰ Filter by time	Time-series analysis
pandas_diff_frames	🔄 Compare datasets	Find differences
pandas_schema_check	✅ Validate schema	Ensure data quality
pandas_sample_stratified	🎲 Stratified sample	Balanced sampling
pandas_sample_random	🎲 Random sample	Quick data preview
pandas_count_distinct	🔢 Count unique values	Cardinality analysis
pandas_export	💾 Export to CSV/Parquet/Excel	Save results
dataset_head_tail	📄 Get first/last N rows	Quick data preview

---

⚡ Polars Tools (3 tools)

Lightning-fast data processing with Polars.

Tool	What it does	When to use
polars_from_rows	📊 Create DataFrame	Build datasets
polars_groupby	🔢 Group + aggregate	Fast aggregations
polars_export	💾 Export to CSV/Parquet/JSON	Save results

---

🦆 DuckDB Tools (1 tool)

SQL on local data sources. No server needed.

Tool	What it does	When to use
duckdb_query_local	🔍 SQL on local data	Query Parquet/CSV files

---

🎨 JSON & Data Quality (5 tools)

Transform, validate, and correlate events.

Tool	What it does	When to use
jq_transform	🔄 JMESPath transform	Extract/transform JSON
event_validate	✅ JSONSchema validation	Validate data structure
event_fingerprint	🔑 Generate fingerprint	Create unique IDs
dedupe_by_id	🧹 Deduplicate records	Remove duplicates
event_correlate	🔗 Correlate events	Join across batches

---

📁 File & Format Tools (4 tools)

Work with files and convert formats.

Tool	What it does	When to use
parquet_inspect	🔍 Inspect Parquet schema	Understand file structure
arrow_convert	🔄 Convert formats	Parquet ↔ CSV ↔ JSON
fs_read	📖 Read file (with offset)	Read large files efficiently
fs_list_dir	📂 List directory	Browse filesystem
rg_search	🔎 Search with ripgrep	Find patterns in code/data

---

📊 Visualization & Evidence (2 tools)

Generate charts and evidence bundles automatically.

Tool	What it does	When to use
dataset_to_chart	📈 Auto-generate charts	Visualize data instantly
evidence_bundle_plus	📦 Generate evidence bundle	Export stats + samples

---

🏗️ Infrastructure Tools (8 tools)

Monitor Nomad jobs and Consul services.

Tool	What it does	When to use
nomad_list_jobs	📋 List Nomad jobs (fuzzy search)	Find running services
nomad_get_job_status	📊 Get full job status	Inspect job details
nomad_get_allocation_logs	📜 Read allocation logs	Debug failing jobs
nomad_get_allocation_events	📅 Get lifecycle events	Track job history
nomad_get_node_status	🖥️ Get node status	Check node health
consul_get_service_ips	🌐 Get service IPs (fuzzy)	Find service endpoints
consul_list_services	📚 List all services	Discover service catalog
consul_get_service_health	❤️ Get service health	Check service status

Fuzzy search: Find services even with partial names! 🎯

---

🌐 HTTP Tools (1 tool)

Make HTTP requests safely.

Tool	What it does	When to use
http_request	🌐 HTTP GET/POST	Call APIs, test endpoints

Safety: POST disabled by default. Opt-in required. ✅

---

✅ Great Expectations (1 tool)

Data quality validation with GE.

Tool	What it does	When to use
great_expectations_check	✅ Run GE validation	Validate data quality

---

🔄 Reconciliation (1 tool)

Compare datasets and find differences.

Tool	What it does	When to use
reconcile_counts	🔍 Reconcile record counts	Compare datasets

---

🔒 Security & Guardrails

🛡️ Built-in Safety

• Filesystem Allowlist: All file operations restricted to MCPKIT_ROOTS
• SQL Validation: JDBC and Athena queries validated (read-only by default)
• HTTP Safety: POST requests disabled by default
• Output Limits: Automatic caps on rows, records, and bytes
• Path Traversal Protection: Blocks .. and unsafe paths

📏 Configurable Limits

Variable	Default	What it does
MCPKIT_TIMEOUT_SECS	15	Operation timeout
MCPKIT_MAX_OUTPUT_BYTES	1000000	Max output size
MCPKIT_MAX_ROWS	500	Max rows returned
MCPKIT_MAX_RECORDS	500	Max records returned

---

⚙️ Configuration

🔐 AWS Credentials

AWS uses the standard boto3 credential chain:

1. Environment Variables (recommended):

   export AWS_ACCESS_KEY_ID=your-key
   export AWS_SECRET_ACCESS_KEY=your-secret
   export AWS_REGION=eu-central-1
   

2. AWS Credentials File (~/.aws/credentials):

   [default]
   aws_access_key_id = your-key
   aws_secret_access_key = your-secret
   

3. IAM Roles (auto-detected on EC2/ECS/Lambda)

4. AWS SSO (use aws sso login first)

📁 Filesystem Security

export MCPKIT_ROOTS="/allowed/path1:/allowed/path2"

🎧 Kafka Configuration

export MCPKIT_KAFKA_BOOTSTRAP="localhost:9092"
export MCPKIT_KAFKA_SECURITY_PROTOCOL="SASL_SSL"  # optional
export MCPKIT_KAFKA_SASL_MECHANISM="PLAIN"  # optional
export MCPKIT_KAFKA_SASL_USERNAME="user"  # optional
export MCPKIT_KAFKA_SASL_PASSWORD="pass"  # optional

📋 Schema Registry

export MCPKIT_SCHEMA_REGISTRY_URL="http://localhost:8081"
export MCPKIT_SCHEMA_REGISTRY_BASIC_AUTH="user:pass"  # optional

🗄️ JDBC

export MCPKIT_JDBC_DRIVER_CLASS="org.postgresql.Driver"
export MCPKIT_JDBC_URL="jdbc:postgresql://localhost/test"
export MCPKIT_JDBC_JARS="/path/to/postgresql.jar"
export MCPKIT_JDBC_USER="user"  # optional
export MCPKIT_JDBC_PASSWORD="pass"  # optional

---

🎯 Example Workflows

🔍 End-to-End Kafka Pipeline Debugging

You: "Debug the 'order-events' topic: consume recent messages, decode Avro, filter errors, analyze patterns, and create a chart"

AI Response:

I'll debug the order-events pipeline end-to-end.

[Step 1: kafka_consume_tail] Consuming last 50 messages...
✓ Consumed 50 messages from 'order-events'

[Step 2: schema_registry_get] Fetching Avro schema...
✓ Schema retrieved: order-event v2.3 (ID: 145)

[Step 3: avro_decode] Decoding Avro messages...
✓ Decoded 50 messages successfully

[Step 4: kafka_filter] Filtering error events...
✓ Found 8 error events matching criteria

[Step 5: pandas_groupby] Analyzing error patterns...
Error breakdown:
- Payment failures: 4 (50%)
- Validation errors: 2 (25%)
- Timeout errors: 2 (25%)

[Step 6: dataset_to_chart] Creating visualization...
✓ Chart saved: .artifacts/order_errors_analysis.png

[Step 7: pandas_export] Exporting error details...
✓ CSV exported: .artifacts/order_errors_20251221.csv

Summary: 8 errors out of 50 messages (16% error rate). Payment failures are the primary issue.

Tools used: kafka_consume_tail → schema_registry_get → avro_decode → kafka_filter → pandas_groupby → dataset_to_chart → pandas_export (7 tools)

---

📊 Multi-Source Data Reconciliation

You: "Compare sales data between Athena and Postgres: query both, join on order_id, find discrepancies, and export differences"

AI Response:

I'll reconcile sales data between Athena and Postgres.

[Step 1: athena_start_query] Querying Athena sales table...
✓ Query started: execution_id abc123

[Step 2: athena_poll_query] Waiting for completion...
✓ Query completed in 3.2s (scanned 1.2 GB)

[Step 3: athena_get_results] Fetching results...
✓ Retrieved 5,234 rows from Athena

[Step 4: pandas_from_rows] Storing Athena data...
✓ Dataset created: athena_sales_20251221

[Step 5: db_query_ro] Querying Postgres sales table...
✓ Retrieved 5,189 rows from Postgres

[Step 6: pandas_from_rows] Storing Postgres data...
✓ Dataset created: postgres_sales_20251221

[Step 7: pandas_join] Joining datasets on order_id...
✓ Joined datasets: 5,180 matching orders

[Step 8: pandas_diff_frames] Finding discrepancies...
Discrepancies found:
- Missing in Postgres: 54 orders
- Missing in Athena: 9 orders
- Amount mismatches: 12 orders
- Total discrepancy value: $3,456.78

[Step 9: pandas_export] Exporting differences...
✓ Differences exported: .artifacts/sales_reconciliation_diff.csv

[Step 10: reconcile_counts] Generating reconciliation report...
✓ Report saved: .artifacts/sales_reconciliation_report.json

Tools used: athena_start_query → athena_poll_query → athena_get_results → pandas_from_rows → db_query_ro → pandas_from_rows → pandas_join → pandas_diff_frames → pandas_export → reconcile_counts (10 tools)

---

🏗️ Service Health Investigation

You: "Investigate the 'payment-service': find it in Nomad, check Consul health, get logs, filter errors, correlate with events, and export report"

AI Response:

I'll investigate the payment-service health comprehensively.

[Step 1: nomad_list_jobs] Finding payment-service...
✓ Found: platform-payment-service-app (status: running)

[Step 2: nomad_get_job_status] Getting job details...
✓ Job running on node: node-abc-123
✓ Allocation: alloc-def-456

[Step 3: consul_get_service_health] Checking Consul health...
✓ Service health: passing (3/3 checks passing)
✓ Service IPs: 10.0.1.45:8080, 10.0.1.46:8080

[Step 4: nomad_get_allocation_logs] Retrieving logs...
✓ Retrieved last 200 log lines

[Step 5: pandas_from_rows] Storing logs as dataset...
✓ Dataset created: payment_service_logs

[Step 6: pandas_filter_query] Filtering error logs...
✓ Found 15 error entries in last hour

[Step 7: kafka_consume_batch] Consuming related events...
✓ Consumed 100 payment events from Kafka

[Step 8: event_correlate] Correlating logs with events...
✓ Correlated 12 error logs with payment events

[Step 9: pandas_groupby] Analyzing error patterns...
Error analysis:
- Database timeouts: 8 (53%)
- External API failures: 4 (27%)
- Validation errors: 3 (20%)

[Step 10: evidence_bundle_plus] Generating evidence bundle...
✓ Bundle created:
  - .artifacts/payment_service_investigation_stats.json
  - .artifacts/payment_service_error_logs.csv
  - .artifacts/payment_service_correlated_events.csv

Summary: Service is healthy but experiencing intermittent database timeouts affecting 8 transactions.

Tools used: nomad_list_jobs → nomad_get_job_status → consul_get_service_health → nomad_get_allocation_logs → pandas_from_rows → pandas_filter_query → kafka_consume_batch → event_correlate → pandas_groupby → evidence_bundle_plus (10 tools)

---

✅ Data Quality Validation Pipeline

You: "Validate Kafka events: consume batch, flatten nested JSON, validate against schema, check data quality, and generate report"

AI Response:

I'll run a complete data quality validation on Kafka events.

[Step 1: kafka_consume_batch] Consuming event batch...
✓ Consumed 500 messages from 'user-actions' topic

[Step 2: kafka_flatten] Flattening nested JSON structures...
✓ Flattened 500 records into 45 columns
✓ Dataset created: user_actions_flattened

[Step 3: schema_registry_get] Fetching validation schema...
✓ Retrieved JSONSchema: user-action-event v1.2

[Step 4: pandas_sample_random] Sampling for validation...
✓ Sampled 50 random records

[Step 5: event_validate] Validating against schema...
Validation results:
- Valid records: 47 (94%)
- Invalid records: 3 (6%)
  - Missing required field 'timestamp': 2
  - Invalid enum value 'action_type': 1

[Step 6: pandas_schema_check] Checking data types and constraints...
✓ Schema check passed: all required columns present
⚠️ Warning: 12% null values in 'metadata' column

[Step 7: pandas_count_distinct] Analyzing data quality...
Distinct value counts:
- user_id: 234 unique (expected: ~250)
- action_type: 8 distinct (expected: 10)
- session_id: 198 unique

[Step 8: pandas_filter_time_range] Filtering recent events...
✓ Filtered to last 24 hours: 487 events

[Step 9: great_expectations_check] Running GE validation suite...
✓ Expectations passed: 8/10
⚠️ Failed: 
  - expect_column_values_to_be_unique (user_id)
  - expect_column_values_to_not_be_null (metadata)

[Step 10: evidence_bundle_plus] Generating quality report...
✓ Quality report: .artifacts/data_quality_report_20251221.zip

Tools used: kafka_consume_batch → kafka_flatten → schema_registry_get → pandas_sample_random → event_validate → pandas_schema_check → pandas_count_distinct → pandas_filter_time_range → great_expectations_check → evidence_bundle_plus (10 tools)

---

📈 Time-Series Analysis & Forecasting

You: "Analyze sales trends: query Athena for last 6 months, filter by date range, group by week, sample for quality, create chart, and export"

AI Response:

I'll analyze sales trends over the last 6 months.

[Step 1: athena_start_query] Starting sales query...
✓ Query: SELECT * FROM sales WHERE date >= '2025-06-01'

[Step 2: athena_poll_query] Monitoring execution...
✓ Query running... (scanned 2.1 GB so far)

[Step 3: athena_get_results] Fetching results...
✓ Retrieved 45,678 sales records

[Step 4: pandas_from_rows] Storing in dataset...
✓ Dataset created: sales_6months

[Step 5: pandas_filter_time_range] Filtering date range...
✓ Filtered: 2025-06-01 to 2025-12-21 (45,234 records)

[Step 6: pandas_groupby] Grouping by week and region...
✓ Aggregated: 26 weeks × 4 regions = 104 groups

[Step 7: pandas_sample_stratified] Stratified sampling...
✓ Sampled 20 records per region for validation

[Step 8: pandas_describe] Statistical analysis...
Statistics:
- Total sales: $12,345,678
- Average per week: $474,833
- Trend: +5.2% week-over-week
- Peak week: Week 24 ($567,890)

[Step 9: dataset_to_chart] Creating time-series chart...
✓ Chart: .artifacts/sales_trends_6months.png

[Step 10: pandas_export] Exporting aggregated data...
✓ Exported: .artifacts/sales_by_week_region.csv (104 rows)
✓ Exported: .artifacts/sales_statistics.json

Tools used: athena_start_query → athena_poll_query → athena_get_results → pandas_from_rows → pandas_filter_time_range → pandas_groupby → pandas_sample_stratified → pandas_describe → dataset_to_chart → pandas_export (10 tools)

---

🔄 Cross-System Event Correlation

You: "Correlate user events across systems: get Kafka events, query database for user profiles, join datasets, deduplicate, correlate timestamps, and export"

AI Response:

I'll correlate user events across Kafka and database systems.

[Step 1: kafka_consume_batch] Consuming user events...
✓ Consumed 1,000 events from 'user-events' topic

[Step 2: kafka_flatten] Flattening event structure...
✓ Flattened to 1,000 records with 32 columns

[Step 3: db_query_ro] Querying user profiles...
✓ Retrieved 850 user profiles from database

[Step 4: pandas_from_rows] Storing profiles...
✓ Dataset created: user_profiles

[Step 5: pandas_join] Joining events with profiles...
✓ Joined on user_id: 987 matched records

[Step 6: dedupe_by_id] Removing duplicate events...
✓ Deduplicated: 23 duplicates removed (964 unique)

[Step 7: event_correlate] Correlating by timestamp...
✓ Correlated events into 234 user sessions

[Step 8: pandas_groupby] Analyzing session patterns...
Session analysis:
- Average session duration: 12.5 minutes
- Events per session: 4.1
- Most active users: 12 users with 10+ events

[Step 9: pandas_filter_query] Filtering high-value sessions...
✓ Found 45 sessions with purchase events

[Step 10: pandas_export] Exporting correlated data...
✓ Exported: .artifacts/correlated_user_sessions.csv
✓ Exported: .artifacts/session_analysis.json

Tools used: kafka_consume_batch → kafka_flatten → db_query_ro → pandas_from_rows → pandas_join → dedupe_by_id → event_correlate → pandas_groupby → pandas_filter_query → pandas_export (10 tools)

---

🧪 Testing

Quick Start

# Run all tests
pytest

# Run unit tests only (no Docker required)
make test-unit

# Run integration tests (requires Docker)
make test-integration

# Run with coverage
make test-coverage

Coverage

# Run tests with coverage
make test-coverage

# Generate HTML report
make coverage-html
open htmlcov/index.html

Target: 95% coverage for core modules. See tests/COVERAGE.md for details.

All tests use Docker containers for external services (Kafka, PostgreSQL, LocalStack, Consul, Nomad). No real AWS credentials needed! ✅

🔄 Reloading MCP Server After Code Changes

When you modify MCP tool code (e.g., mcpkit/core/*.py or mcpkit/server.py), Cursor needs to reload the MCP server to pick up your changes.

Quick reload:

./reload_mcp_server.sh

This script modifies your Cursor MCP config file to trigger an automatic reload. If automatic reload doesn't work, restart Cursor manually.

When to reload:

• ✅ After editing tool implementations in mcpkit/core/*.py
• ✅ After adding/modifying tools in mcpkit/server.py
• ✅ After changing tool documentation or parameters
• ❌ Not needed for config changes (environment variables)

Manual reload alternative:

1. Open Cursor settings → MCP
2. Temporarily disable the mcpkit-data server
3. Re-enable it
4. Cursor will reload the server

---

🤝 Contributing

We ❤️ contributions! Here's how to help:

1. 🐛 Found a bug? Open an issue
2. 💡 Have an idea? Propose a new tool or feature
3. 🔧 Want to code? Check out TOOLS_ANALYSIS.md for gaps
4. 📝 Docs unclear? Improve them!

Design Principles:

• 🎯 KISS: One tool, one job
• 🔗 Composable: Tools work together seamlessly
• 🔒 Safe: Read-only by default, explicit opt-ins
• ⚡ Fast: Efficient operations, smart limits

---

📚 Learn More

• Model Context Protocol - The protocol we use
• FastMCP - The framework we build on
• Cursor - The IDE that makes this shine ✨

---

📄 License

MIT License - Use it, fork it, make it better! 🚀

---

Made with ❤️ for data engineers who want to ship faster

Stop context-switching. Start shipping. 🎯