hindsight-mcp
Consolidates git history, test results, and Copilot sessions into a searchable SQLite database for AI assistants to answer questions about development activity.
README
<img src="https://raw.githubusercontent.com/Rbfinch/hindsight-mcp/main/assets/hindsight-logo.png" alt="Hindsight Logo" width="120">
<!-- mcp-name: io.github.Rbfinch/hindsight-mcp -->
hindsight-mcp
An MCP server for AI-assisted coding that leverages development history.
Overview
hindsight-mcp consolidates development data (git logs, test results, and GitHub Copilot sessions) into a searchable SQLite database, making it accessible to AI assistants through MCP tool calls in VS Code.
Key Features:
- Full-text search across commits and Copilot conversations
- Track test results linked to specific commits
- Activity summaries and timeline views
- Automatic git and Copilot session ingestion
Quick Start
Prerequisites
- VS Code v1.99+ with GitHub Copilot
- cargo-nextest (for test ingestion):
cargo install cargo-nextest
Installation
cargo install hindsight-mcp
Configure VS Code
Create .vscode/mcp.json in your project:
{
"servers": {
"hindsight": {
"type": "stdio",
"command": "hindsight-mcp",
"args": ["--workspace", "${workspaceFolder}"]
}
}
}
Verify Setup
- Open VS Code Command Palette (
Cmd+Shift+P) - Run "MCP: List Servers"
- Confirm
hindsightis listed - In Copilot Chat, switch to Agent mode
- Ask: "What have I been working on recently?"
That's it! Copilot will use hindsight tools to answer questions about your development history.
MCP Tools
| Tool | Purpose | Example Prompt |
|---|---|---|
hindsight_timeline |
Chronological activity view | "Show recent commits and test runs" |
hindsight_search |
Full-text search | "Find commits about authentication" |
hindsight_failing_tests |
Query test failures | "What tests are failing?" |
hindsight_activity_summary |
Aggregate stats | "Summarise my week" |
hindsight_commit_details |
Commit info with tests | "Details for commit abc123" |
hindsight_ingest |
Trigger data refresh | "Refresh development history" |
<details> <summary><strong>Tool Arguments Reference</strong></summary>
hindsight_timeline
limit(int): Max events, default 50workspace(string): Filter by path
hindsight_search
query(string): Search query (required)source(string): "all", "commits", or "messages"limit(int): Max results, default 20
hindsight_failing_tests
limit(int): Max tests, default 50workspace(string): Filter by pathcommit(string): Filter by SHA
hindsight_activity_summary
days(int): Days to summarise, default 7
hindsight_commit_details
sha(string): Commit SHA (required)
hindsight_ingest
workspace(string): Path to ingest (required)source(string): "git", "copilot", or "all"incremental(bool): Only new data, default truelimit(int): Max items
</details>
Test Ingestion
Run tests and automatically ingest results:
# Run all tests and ingest
hindsight-mcp test
# Test specific package
hindsight-mcp test -p my-crate
# Preview without writing to database
hindsight-mcp test --dry-run
The test command automatically:
- Spawns
cargo nextestwith correct flags - Auto-detects the current git commit
- Ingests results to the database
<details> <summary><strong>CI / Advanced Usage</strong></summary>
For CI pipelines or custom nextest invocations:
# Using stdin mode
NEXTEST_EXPERIMENTAL_LIBTEST_JSON=1 cargo nextest run \
--message-format libtest-json 2>/dev/null | \
hindsight-mcp test --stdin
# Using ingest command with explicit commit
NEXTEST_EXPERIMENTAL_LIBTEST_JSON=1 cargo nextest run \
--message-format libtest-json 2>/dev/null | \
hindsight-mcp ingest --tests --commit $(git rev-parse HEAD)
</details>
Data Sources
| Source | Data Collected |
|---|---|
| Git | Commits (SHA, author, message, timestamp, parents) |
| Tests | Run metadata, outcomes, durations, failure output |
| Copilot | Chat sessions, prompts, responses, attached files |
Git and Copilot data are ingested automatically. Test results require running hindsight-mcp test.
CLI Reference
hindsight-mcp [OPTIONS] [COMMAND]
Commands:
ingest Ingest data from various sources
test Run tests and ingest results
Options:
-d, --database <PATH> Database path [default: ~/.hindsight/hindsight.db]
-w, --workspace <PATH> Workspace path [default: current directory]
-v, --verbose Debug logging
-q, --quiet Errors only
--skip-init Skip database init
-h, --help Print help
-V, --version Print version
<details> <summary><strong>Test Subcommand Options</strong></summary>
hindsight-mcp test [OPTIONS] [-- <NEXTEST_ARGS>...]
Options:
-p, --package <PKG> Package(s) to test
--bin <BIN> Binary(ies) to run
-E, --filter <EXPR> Filter expression
--stdin Read from stdin
--dry-run Preview only
--no-commit Do not link to commit
--commit <SHA> Explicit commit SHA
--show-output Show test output
</details>
Environment Variables
| Variable | Description |
|---|---|
HINDSIGHT_DATABASE |
Database path |
HINDSIGHT_WORKSPACE |
Default workspace |
Database Location
| Platform | Default Path |
|---|---|
| macOS | ~/Library/Application Support/hindsight/hindsight.db |
| Linux | ~/.local/share/hindsight/hindsight.db |
| Windows | %LOCALAPPDATA%\hindsight\hindsight.db |
Troubleshooting
| Problem | Solution |
|---|---|
| Server does not start | Check binary path; run with --verbose |
| No data showing | Run hindsight_ingest tool via Copilot |
| Log interference | Use --quiet in production |
Development
<details> <summary><strong>Building from Source</strong></summary>
git clone https://github.com/Rbfinch/hindsight-mcp.git
cd hindsight-mcp
cargo build --release
</details>
<details> <summary><strong>Running Tests</strong></summary>
cargo install cargo-nextest
cargo nextest run --workspace
</details>
<details> <summary><strong>Workspace Structure</strong></summary>
hindsight/
├── crates/
│ ├── hindsight-mcp/ # MCP server binary
│ ├── hindsight-git/ # Git log processing
│ ├── hindsight-tests/ # Test result processing
│ └── hindsight-copilot/ # Copilot session parsing
</details>
<details> <summary><strong>Fuzzing</strong></summary>
cd crates/hindsight-tests
cargo +nightly fuzz run fuzz_nextest_run
cd crates/hindsight-copilot
cargo +nightly fuzz run fuzz_session_json
</details>
License
MIT
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。