# Detecting Provider

> Detects the active data source provider and retrieves configuration (database IDs, constants). Internal shared skill — not for direct user invocation.

- Skill: `kazukinagata/detecting-provider` (Agent Skill)
- Install (CLI): `npx skillmds@latest add kazukinagata/detecting-provider`
- Raw SKILL.md: https://api.skillmd.com/api/skills/kazukinagata/detecting-provider/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: kazukinagata (https://skillmd.com/u/kazukinagata)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/kazukinagata/detecting-provider

---


# Agentic Tasks — Provider Detection

Determine the active provider using the following layered check.
**Skip if already determined in this conversation.**

## Layer 1: MCP Tool Auto-Detection
Inspect which MCP tools are available:
- `notion-*` tools present → active_provider = **notion**
- `mcp__airtable__*` tools present → active_provider = **airtable**
- SQLite/database tools present → active_provider = **sqlite**

If exactly one provider MCP is detected, use it.

**REQUIRED — Read `${CLAUDE_PLUGIN_ROOT}/skills/providers/{active_provider}/SKILL.md` now.**
This file contains query method selection logic (Query Path Detection) and data operation procedures.
Skipping this step causes incorrect query paths and suboptimal performance.
Do NOT use MCP search/fetch tools for task queries until you have read the provider SKILL.md.

## Layer 2: Conflict Resolution (multiple provider MCPs detected)
If multiple provider MCPs are detected, determine the environment:
- **Claude Code**: Check `env.AGENTIC_TASKS_PROVIDER` in `~/.claude/settings.json`
- **Cowork / Global Instructions**: Look for `AGENTIC_TASKS_PROVIDER: <value>` in the Global Instructions or CLAUDE.md

If a value is found, use it as active_provider. **REQUIRED — Read the corresponding provider SKILL.md** (same instruction as Layer 1).

## Layer 3: Ask User
If provider is still undetermined, use AskUserQuestion:
> "Multiple data source MCPs are available. Which provider should I use for agentic-tasks? Available: [list detected providers]"

## No MCP Detected
If no provider MCP is found at all, inform the user they need to run the **setting-up-tasks** skill first to configure a data source, then stop.

## Environment Detection

After detecting the provider, also determine the execution environment and set `execution_environment` as a conversation context variable.
**Skip if already set in this conversation.**

Detection logic:
1. If environment variable `CLAUDE_CODE_IS_COWORK` is `1` → `execution_environment = "cowork"`
2. Otherwise → `execution_environment = "claude-code"`

(`CLAUDECODE=1` is common to both environments, so it is not used for detection.)

This value is used by downstream skills (executing-tasks, managing-tasks, etc.) for execution flow branching.

## Config Retrieval

After detecting the provider, retrieve database IDs and constants from the Config page.
**Skip if `headless_config` is already set in this conversation.**

Follow the active provider SKILL.md's Config Retrieval section to populate `headless_config`.

## Constants

Constants shared across skills. All skills that go through detecting-provider reference these values.

| Constant | Value | Purpose |
|--------|-----|------|
| `maxConcurrentAgents` | 3 | Default maximum parallel agents (configurable in headless_config) |