Sling - Data Integration Platform

Move data between 40+ databases, file systems, and APIs with a single CLI.

When to Use Sling

MCP Tools Reference

The Sling CLI stdio MCP has no run, build, or report action. Execute with the CLI: sling run (replications/pipelines), sling build (SQL models), and sling assist report (issue reports). The Platform HTTP MCP is separate; its jobs.trigger runs hosted jobs, not local files.

After you classify a failure as a sling defect (not user config), offer: "Report this? GitHub issue (public, needs account) or email to support." Never offer this for user-config errors. Send only after the user agrees. Show the user the draft first (sling assist report --id <id>), then send with --github or --email. Add --submit to skip the interactive confirm (for agents and non-TTY use). --email opens the contact page with the report prefilled; the user fills in their name and email there — do not ask for an email address yourself.

Improve the report before sending: pass --title "<plain-English summary>" (the default title is a terse error label, e.g. did_not_find_any_columns_for_ident_ident; write what a maintainer would search for, e.g. "Incremental replication fails when source table has no columns") and --description "<context>" (what you tried, what you expected, what happened; it appears above the raw error). Example:

sling assist report --id <id> --github --submit \
  --title "Incremental replication fails when source table has no columns" \
  --description "Replication from Postgres to ClickHouse fails on the first stream. The source table exists and the user can select from it."

Quick Examples

// List connections
{"action": "list", "input": {}}

// Test connection
{"action": "test", "input": {"connection": "MY_CONN", "debug": true}}

// Validate a replication (compile is the default)
{"action": "validate", "input": {"file_path": "/path/to/replication.yaml"}}

CLI vs MCP — which to use

Use the MCP tools to inspect and validate. Use the CLI to execute.

Do not query a database with sling run --stdout. That flag is for piping data to a target, not for inspecting a database. Use the MCP database tool.

When the MCP tools are not available, query with sling conns exec:

sling conns exec MY_PG "select * from public.users limit 10"
sling conns exec MY_PG "select 1" -o json      # Output: text (default), csv, json, arrow

CLI Quick Reference

sling conns list                    # List connections
sling conns test MY_CONN --debug    # Test connection
sling conns discover MY_PG          # Discover tables
sling conns exec MY_PG "select 1"   # Run a SQL query (add -o json|csv|arrow)
sling run -r replication.yaml       # Run replication
sling run -p pipeline.yaml          # Run pipeline
sling run -r replication.yaml --debug   # Run with debug logging
sling build run                     # Run SQL models
sling build compile                 # Compile models without executing
sling assist --id <id>              # Investigate a failure (id from the error footer)
sling assist error <sig>            # Look up an error signature (from prompt context)
sling assist report --id <id>       # Review a redacted issue report
sling assist                        # Open assist (offers to investigate failures)

Topics Reference

Documentation Map

Append .md paths to https://docs.slingdata.io to fetch raw markdown. Full index: https://docs.slingdata.io/llms.txt

Documentation

• Official Docs: https://docs.slingdata.io
• GitHub: https://github.com/slingdata-io/sling-cli
• Discord: https://discord.gg/q5xtaSNDvp