{"title":"Claude Code 中的 MCP","slug":"claude-code","category":"mcp","summary":"在 Claude Code 终端 Agent 中使用 Synapse 工具 — 为编码会话提供持久化记忆。","audience":["human","llm"],"tags":["mcp","claude-code","terminal","coding"],"difficulty":"intermediate","updated":"2026-06-27","word_count":122,"read_minutes":1,"llm_context":"Claude Code config: ~/.claude/config.json (or via `claude mcp add` command)\nCommand: npx -y synapse-mcp-api@latest\nEnv: SYNAPSE_MIND_KEY (required), SYNAPSE_URL (optional)\nAlternative: `claude mcp add synapse -- npx -y synapse-mcp-api@latest`\nThen: set SYNAPSE_MIND_KEY env var in your shell\nTest: in claude code, type \"recall all my memories\"\n","lang":"zh","translated":true,"requested_lang":"zh","content_markdown":"\n# Claude Code 中的 MCP\n\nClaude Code 是 Anthropic 推出的基于终端的编码 Agent。借助 Synapse MCP，Claude Code 在编码会话之间获得持久化记忆 — 它会记住你的项目上下文、过往决策和代码库模式。\n\n## 配置\n\n### 方法 1：CLI 命令（推荐）\n\n```bash\n# 把 Synapse 添加为 MCP Server\nclaude mcp add synapse -- npx -y synapse-mcp-api@latest\n\n# 把 Mind Key 设为环境变量（添加到 ~/.bashrc 或 ~/.zshrc）\nexport SYNAPSE_MIND_KEY=\"mk_YOUR_MIND_KEY_HERE\"\nexport SYNAPSE_URL=\"https://synapse.schaefer.zone\" # 可选，默认为 synapse.schaefer.zone\n```\n\n### 方法 2：编辑配置文件\n\n编辑 `~/.claude/config.json`：\n\n```json\n{\n \"mcpServers\": {\n \"synapse\": {\n \"command\": \"npx\",\n \"args\": [\"-y\", \"synapse-mcp-api@latest\"],\n \"env\": {\n \"SYNAPSE_MIND_KEY\": \"mk_YOUR_MIND_KEY_HERE\",\n \"SYNAPSE_URL\": \"https://synapse.schaefer.zone\"\n }\n }\n }\n}\n```\n\n## 验证可用\n\n启动 Claude Code：\n\n```bash\nclaude\n```\n\n在 Claude Code 提示符中输入：\n\n```\nrecall all my memories\n```\n\nClaude 会调用 `memory_recall` 并返回你已存储的记忆。\n\n## 常见模式\n\n### 项目上下文持久化\n\n编码会话开始时：\n\n```\nrecalling my memories — what projects am I working on?\n```\n\nClaude 调用 `memory_recall`，查看你的项目列表，并从中断处继续工作。\n\n### 代码库决策\n\n当你做出架构决策时：\n\n```\nstore this: \"We chose PostgreSQL over SQLite because of concurrent write\nperformance. Decision made on 2026-06-27.\"\n```\n\nClaude 调用 `memory_store`，category 为 `project`，priority 为 `high`。\n\n### 避免过往错误\n\n```\nwhat mistakes have I made with this codebase?\n```\n\nClaude 搜索 category 为 `mistake` 的记忆并提醒你过往错误。\n\n### 任务跟踪\n\n```\ncreate a task: \"Refactor auth module to use JWT refresh tokens\"\n```\n\nClaude 调用 `task_create`，任务跨会话持久化。\n\n## 工具配置文件\n\n当你不需要全部 119 个工具时：\n\n```bash\n# 通过环境变量设置 profile\nexport MCP_PROFILE=\"standard\" # 25 个工具，约 2500 tokens\n\n# 或在 config.json 中\n\"env\": {\n \"SYNAPSE_MIND_KEY\": \"mk_...\",\n \"MCP_PROFILE\": \"standard\"\n}\n```\n\n## 故障排查\n\n### MCP Server 未启动\n\n```bash\n# 手动测试 MCP Server\nnpx -y synapse-mcp-api@latest\n\n# 应输出：\n# [info] synapse-mcp Streamable HTTP server ready\n# [info] MCP: http://localhost:13100/mcp\n```\n\n### Mind Key 未识别\n\n```bash\n# 确认环境变量已设置\necho $SYNAPSE_MIND_KEY\n\n# 直接测试\ncurl -H \"Authorization: Bearer $SYNAPSE_MIND_KEY\" \\\n https://synapse.schaefer.zone/memory/recall\n```\n\n### Claude Code 看不到工具\n\n- 配置更改后重启 Claude Code\n- 检查 `claude mcp list` 确认 Synapse 已注册\n- 参见 [MCP 故障排查](/docs/mcp/troubleshooting)\n\n## 下一步\n\n- [Claude Desktop 配置](/docs/mcp/claude-desktop)\n- [Cursor 配置](/docs/mcp/cursor)\n- [持久化 LLM Agent 指南](/docs/guides/persistent-llm-agent)\n","content_html":"

Claude Code 中的 MCP

\n

Claude Code 是 Anthropic 推出的基于终端的编码 Agent。借助 Synapse MCP，Claude Code 在编码会话之间获得持久化记忆 — 它会记住你的项目上下文、过往决策和代码库模式。

\n

配置

\n

方法 1：CLI 命令（推荐）

\n
# 把 Synapse 添加为 MCP Server\nclaude mcp add synapse -- npx -y synapse-mcp-api@latest\n\n# 把 Mind Key 设为环境变量（添加到 ~/.bashrc 或 ~/.zshrc）\nexport SYNAPSE_MIND_KEY="mk_YOUR_MIND_KEY_HERE"\nexport SYNAPSE_URL="https://synapse.schaefer.zone"  # 可选，默认为 synapse.schaefer.zone

方法 2：编辑配置文件

\n

编辑 ~/.claude/config.json：

\n
{\n  "mcpServers": {\n    "synapse": {\n      "command": "npx",\n      "args": ["-y", "synapse-mcp-api@latest"],\n      "env": {\n        "SYNAPSE_MIND_KEY": "mk_YOUR_MIND_KEY_HERE",\n        "SYNAPSE_URL": "https://synapse.schaefer.zone"\n      }\n    }\n  }\n}

验证可用

\n

启动 Claude Code：

\n
claude

在 Claude Code 提示符中输入：

\n
recall all my memories

Claude 会调用 memory_recall 并返回你已存储的记忆。

\n

常见模式

\n

项目上下文持久化

\n

编码会话开始时：

\n
recalling my memories — what projects am I working on?

Claude 调用 memory_recall，查看你的项目列表，并从中断处继续工作。

\n

代码库决策

\n

当你做出架构决策时：

\n
store this: "We chose PostgreSQL over SQLite because of concurrent write\nperformance. Decision made on 2026-06-27."

Claude 调用 memory_store，category 为 project，priority 为 high。

\n

避免过往错误

\n
what mistakes have I made with this codebase?

Claude 搜索 category 为 mistake 的记忆并提醒你过往错误。

\n

任务跟踪

\n
create a task: "Refactor auth module to use JWT refresh tokens"

Claude 调用 task_create，任务跨会话持久化。

\n

工具配置文件

\n

当你不需要全部 119 个工具时：

\n
# 通过环境变量设置 profile\nexport MCP_PROFILE="standard"  # 25 个工具，约 2500 tokens\n\n# 或在 config.json 中\n"env": {\n  "SYNAPSE_MIND_KEY": "mk_...",\n  "MCP_PROFILE": "standard"\n}

故障排查

\n

MCP Server 未启动

\n
# 手动测试 MCP Server\nnpx -y synapse-mcp-api@latest\n\n# 应输出：\n# [info] synapse-mcp Streamable HTTP server ready\n# [info]   MCP:      http://localhost:13100/mcp

Mind Key 未识别

\n
# 确认环境变量已设置\necho $SYNAPSE_MIND_KEY\n\n# 直接测试\ncurl -H "Authorization: Bearer $SYNAPSE_MIND_KEY" \\\n     https://synapse.schaefer.zone/memory/recall

Claude Code 看不到工具

\n\n

下一步

\n\n","urls":{"html":"/docs/mcp/claude-code","text":"/docs/mcp/claude-code?format=text","json":"/docs/mcp/claude-code?format=json","llm":"/docs/mcp/claude-code?format=llm"},"translations_available":["en","zh","hi","es","fr","ar","pt","ru","ja","de","it","ko","nl","pl","tr","sv","vi","th","id","uk"]}