Compatibility
Claude Code × DeepSeek
Partial Support5 runs · Success rate 80% · Last verified 2026-08-18
Verification
- Source Verified
- Install Verified
- Run Verified
- Security Reviewed
Capabilities
| Capabilities | Result |
|---|---|
| Installation | Not tested |
| Authentication | Not tested |
| Chat | Not tested |
| Tool Calling | Works |
| File Editing | Works |
| MCP | Not tested |
| Agent Skills | Not tested |
Tested Environment
- Agent version
- 2.0.38
- Target version
- deepseek-chat
- Test date
- 2026-08-18
How We Tested
Task
Expected outcome
Test Results
5 runs · 4 passed · 1 failed
| Date | Status | Duration | Tokens | Cost | |
|---|---|---|---|---|---|
| 2026-08-18 | Passed | 1.2s | — | — | View run → |
| 2026-08-18 | Passed | 0ms | — | — | View run → |
| 2026-08-16 | Failed | 45.0s | 8,210 | $0.09 | View run → |
| 2026-08-15 | Passed | 1m 21s | 13,980 | $0.21 | View run → |
| 2026-08-14 | Passed | 1m 12s | 12,450 | $0.18 | View run → |
Known Issues
Tool-calling is unreliable with deepseek-reasoner
In our runs deepseek-reasoner returned malformed JSON arguments for parallel tool calls.
workaround: Use deepseek-chat for tool-heavy workflows.
since 2026-08-16
Working Configuration
# Claude Code × DeepSeek (verified by AgentRunDB runs) export ANTHROPIC_BASE_URL=https://api.deepseek.com export ANTHROPIC_AUTH_TOKEN=$DEEPSEEK_API_KEY # model: deepseek-chat (tool calling verified) — avoid deepseek-reasoner for tool-heavy tasks
Recipes
Reproducible configurations backed by these runs. /recipes →
Compared With
Side-by-side comparisons for this target. /compare →