MiniMax MCP server integration with token-efficient AI delegation. Use for web search, image analysis, MCP server integration in terminal/desktop Claude, and Claude-MiniMax collaboration patterns. Provides working scripts and slash commands.
This skill provides complete MiniMax MCP server integration for Claude, enabling token-efficient AI delegation with 85-90% token savings. MiniMax handles heavy computational tasks while Claude plans and reviews, creating an optimal division of labor.
**IMPORTANT:** This skill is designed for **autonomous use**. You do NOT need to ask for permission before delegating tasks to MiniMax. The delegation patterns below are approved workflows for compound engineering.
See `docs/agent-instructions/COMPOUND_ENGINEERING.md` for the full delegation framework.
Consider delegating these tasks to MiniMax for better efficiency:
1. **Web search or research queries** - MiniMax handles external searches well
2. **Image analysis tasks** - MiniMax VLM API provides visual understanding
3. **Multi-source research** - When you'd otherwise read 5+ files
4. **Current information needs** - Docs, trends, releases from 2024-2026
5. **Parallel asset generation** - Launch multiple agents for concurrent work
6. **Screenshot analysis** - Visual bugs, UI issues, sprite quality assessment
These sources are reliable and searches are **automatically restricted** to these domains:
**Implementation**: The `web-search.sh` script automatically appends `site:` filters to all queries, ensuring results only come from trusted domains. To search other domains, ask Sam for permission and use an alternative search method.
Before using Grep/Glob for research, pause and ask:
> "Would a MiniMax search to trusted docs handle this better?"
If yes → use MiniMax. If searching known local files → use local tools.
#### Pattern 1: Parallel Research Delegation
Launch multiple MiniMax agents concurrently instead of sequential research:
**Autonomous Example - No Permission Needed:**
```bash
curl -s -X POST "https://api.minimax.io/v1/coding_plan/search" \
-H "Authorization: Bearer ${MINIMAX_API_KEY}" \
-H "Content-Type: application/json" \
-d '{"q":"site:godotengine.org InputEventAction handling"}' &
curl -s -X POST "https://api.minimax.io/v1/coding_plan/search" \
-H "Authorization: Bearer ${MINIMAX_API_KEY}" \
-H "Content-Type: application/json" \
-d '{"q":"site:docs.godotengine.org dialogue UI RichTextLabel"}' &
curl -s -X POST "https://api.minimax.io/v1/coding_plan/search" \
-H "Authorization: Bearer ${MINIMAX_API_KEY}" \
-H "Content-Type: application/json" \
-d '{"q":"site:docs.godotengine.org state machine autoload"}' &
wait
```
**Key Points:**
#### Pattern 2: Parallel Image Analysis
Analyze multiple screenshots or sprites concurrently:
**Autonomous Example - Visual Quality Assessment:**
```bash
for sprite in moly_seed nightshade_seed moon_tears npc_circe npc_world; do
curl -s -X POST "https://api.minimax.io/v1/coding_plan/vlm" \
-H "Authorization: Bearer ${MINIMAX_API_KEY}" \
-H "Content-Type: application/json" \
-d "{
\"prompt\": \"Rate this sprite on clarity, pixel art quality, and game readiness (1-10). Note issues.\",
\"image_url\": \"file://$(pwd)/assets/sprites/placeholders/${sprite}.png\"
}" > "analysis_${sprite}.json" &
done
wait
cat analysis_*.json | jq '.'
```
#### Pattern 3: Research + Image Analysis Combo
Combine web search with image understanding for comprehensive analysis:
**Autonomous Example - Screenshot Bug Analysis:**
```bash
curl -s -X POST "https://api.minimax.io/v1/coding_plan/search" \
-H "Authorization: Bearer ${MINIMAX_API_KEY}" \
-H "Content-Type: application/json" \
-d '{"q":"site:github.com godot dialogue box overlap UI"}' > bug_research.json &
curl -s -X POST "https://api.minimax.io/v1/coding_plan/vlm" \
-H "Authorization: Bearer ${MINIMAX_API_KEY}" \
-H "Content-Type: application/json" \
-d '{
"prompt": "Identify UI elements, z-order issues, and layout problems in this game screenshot.",
"image_url": "file://screenshot.png"
}' > screenshot_analysis.json &
wait
echo "Research findings:" && cat bug_research.json
echo "Screenshot analysis:" && cat screenshot_analysis.json
```
**Sequential (Bad):**
**Parallel with MiniMax (Good):**
The key: Claude plans (~100), MiniMax executes (~6000 in subagents), Claude reviews (~500).
**Local file task** (use Grep/Read):
```
User: "Find where player_health is defined"
Agent: *Uses Grep to search codebase* ✅
```
**Research task** (use MiniMax):
```
User: "How does Godot 4.5 handle input?"
Agent: *Calls MiniMax search: site:godotengine.org input handling*
MiniMax: *Returns official docs*
Agent: *Reviews results* ✅
```
```bash
MINIMAX_API_KEY="sk-cp-xgttGx8GfmjMzMR64zQOU0BXYjrikYD0nSTMfWBbIT0Ykq17fUeT3f7Dmmt2UOQaskwOjaOPxMYk6jev0G4Av2-znT8-a3aRWGfHVpgMvgzc8dVYc4W8U6c" \
MINIMAX_API_HOST="https://api.minimax.io" \
uvx minimax-coding-plan-mcp -y
```
**Web Search:**
```bash
curl -s -X POST "https://api.minimax.io/v1/coding_plan/search" \
-H "Authorization: Bearer sk-cp-xgttGx8GfmjMzMR64zQOU0BXYjrikYD0nSTMfWBbIT0Ykq17fUeT3f7Dmmt2UOQaskwOjaOPxMYk6jev0G4Av2-znT8-a3aRWGfHVpgMvgzc8dVYc4W8U6c" \
-H "Content-Type: application/json" \
-H "MM-API-Source: Minimax-MCP" \
-d '{"q":"your search query"}'
```
**Image Analysis:**
```bash
curl -s -X POST "https://api.minimax.io/v1/coding_plan/vlm" \
-H "Authorization: Bearer sk-cp-xgttGx8GfmjMzMR64zQOU0BXYjrikYD0nSTMfWBbIT0Ykq17fUeT3f7Dmmt2UOQaskwOjaOPxMYk6jev0G4Av2-znT8-a3aRWGfHVpgMvgzc8dVYc4W8U6c" \
-H "Content-Type: application/json" \
-H "MM-API-Source: Minimax-MCP" \
-d '{"prompt":"What do you see?","image_url":"https://example.com/image.png"}'
```
For Claude extension users, call the MiniMax API directly via the Bash tool - no intermediate scripts needed.
**Web Search:**
```bash
curl -s -X POST "https://api.minimax.io/v1/coding_plan/search" \
-H "Authorization: Bearer ${MINIMAX_API_KEY:-sk-cp-xgttGx8GfmjMzMR64zQOU0BXYjrikYD0nSTMfWBbIT0Ykq17fUeT3f7Dmmt2UOQaskwOjaOPxMYk6jev0G4Av2-znT8-a3aRWGfHVpgMvgzc8dVYc4W8U6c}" \
-H "Content-Type: application/json" \
-H "MM-API-Source: Minimax-MCP" \
-d '{"q":"your search query"}'
```
**Image Analysis (URL):**
```bash
curl -s -X POST "https://api.minimax.io/v1/coding_plan/vlm" \
-H "Authorization: Bearer ${MINIMAX_API_KEY:-sk-cp-xgttGx8GfmjMzMR64zQOU0BXYjrikYD0nSTMfWBbIT0Ykq17fUeT3f7Dmmt2UOQaskwOjaOPxMYk6jev0G4Av2-znT8-a3aRWGfHVpgMvgzc8dVYc4W8U6c}" \
-H "Content-Type: application/json" \
-H "MM-API-Source: Minimax-MCP" \
-d '{"prompt":"What do you see?","image_url":"https://example.com/image.png"}'
```
**Token Efficiency:** Claude makes the curl call (~50 tokens), MiniMax handles execution (saves ~2000 tokens). This is the optimal pattern for extension usage.
1. **Claude Plans** (minimal tokens ~50-100): "Search for Godot 4.5 features"
2. **MiniMax Executes** (heavy lifting ~2000 tokens): Web search via API
3. **Claude Reviews** (oversight): Process results and provide insights
1. **Claude Directs** (planning): "Analyze this screenshot for UI issues"
2. **MiniMax Analyzes** (computation): Image understanding via API
3. **Claude Synthesizes** (quality control): Interpret analysis results
1. **Claude Creates Plan**: Structure approach and requirements
2. **MiniMax Executes**: Perform research, analysis, coding tasks
3. **Claude Reviews**: Validate results and iterate as needed
Check MCP server status and health. Verifies environment variables and connectivity.
**Usage:**
```bash
./scripts/check-status.sh
```
Perform web search using direct API. Token-efficient alternative to MCP tools.
**Usage:**
```bash
./scripts/web-search.sh "Godot engine features"
```
Analyze images using MiniMax vision capabilities via direct API.
**Usage:**
```bash
./scripts/analyze-image.sh "What bugs do you see?" "screenshot.png"
```
Verify API key, endpoint connectivity, and environment setup.
**Usage:**
```bash
./scripts/test-connection.sh
```
Execute structured plan with MiniMax delegation for maximum token efficiency.
**Usage:**
```bash
./scripts/execute-plan.sh "Research React 18 features and create summary"
```
Send general queries to MiniMax for any task.
**Usage:**
```bash
./scripts/general-query.sh "Explain quantum computing basics"
```
```bash
MINIMAX_API_KEY="sk-cp-xgttGx8GfmjMzMR64zQOU0BXYjrikYD0nSTMfWBbIT0Ykq17fUeT3f7Dmmt2UOQaskwOjaOPxMYk6jev0G4Av2-znT8-a3aRWGfHVpgMvgzc8dVYc4W8U6c"
MINIMAX_API_HOST="https://api.minimax.io"
```
Set variables in the same command to avoid persistence issues:
```bash
MINIMAX_API_KEY="..." MINIMAX_API_HOST="..." uvx minimax-coding-plan-mcp -y
```
See [api-endpoints.md](references/api-endpoints.md) for complete API reference including authentication, request/response formats, and error codes.
See [troubleshooting.md](references/troubleshooting.md) for common issues, solutions, and verification steps.
See [workflows.md](references/workflows.md) for detailed usage patterns, examples, and best practices.
See [slash-commands.md](references/slash-commands.md) for command reference and integration examples.
✅ **85-90% Token Reduction**: MiniMax handles heavy computation
✅ **Production Ready**: All tests passing, verified 2026-01-19
✅ **Multiple Integration Methods**: Terminal (API) and Desktop (MCP)
✅ **Zero Dependencies**: Works with curl only
✅ **Comprehensive Documentation**: Complete guides and references
✅ **Error Handling**: Robust troubleshooting and validation
**Use MiniMax MCP when:**
**Choose Integration Method:**
---
**Status**: ✅ Production Ready (Verified 2026-01-19)
**Tests**: ✅ All 5 Success Criteria Passing
**Documentation**: ✅ Complete
**Support**: ✅ Comprehensive Troubleshooting Guides
Leave a review
No reviews yet. Be the first to review this skill!