g-gremlin-hubspot-mcp
Enables advanced HubSpot CRM operations including unlimited record extraction, duplicate detection and merging, property drift analysis, and bulk upserts, surpassing official API limits.
README
g-gremlin-hubspot-mcp
Status: Public Beta — 14 tools shipping, core workflows stable, feedback welcome. See Known Gaps below.
The HubSpot MCP server for teams that hit the API ceiling.
HubSpot's official MCP gives you read-only search capped at 10k records. This one gives you dedup merge plans, auto-windowing past the ceiling, property drift detection, and bulk upserts — all from Claude Desktop, Cursor, or Windsurf.
Powered by g-gremlin, the CLI for Google Workspace and CRM automation.
Quickstart
# 1. Install
pipx install g-gremlin
pipx install g-gremlin-hubspot-mcp
# 2. Connect to HubSpot (one-time)
g-gremlin hubspot connect --access-token YOUR_PRIVATE_APP_TOKEN
# 3. Add to your MCP client
Claude Desktop
Add to your claude_desktop_config.json:
{
"mcpServers": {
"g-gremlin-hubspot": {
"command": "g-gremlin-hubspot-mcp",
"env": {
"G_GREMLIN_HUBSPOT_ACCESS_TOKEN": "YOUR_TOKEN"
}
}
}
}
Cursor / Windsurf
Add to your MCP settings:
{
"mcpServers": {
"g-gremlin-hubspot": {
"command": "g-gremlin-hubspot-mcp"
}
}
}
OpenClaw (community-supported via mcporter)
OpenClaw can call MCP servers through mcporter.
- Install mcporter:
npm install -g mcporter
- Add this MCP server in your mcporter config:
{
"mcpServers": {
"g-gremlin-hubspot": {
"command": "g-gremlin-hubspot-mcp",
"env": {
"G_GREMLIN_HUBSPOT_ACCESS_TOKEN": "YOUR_TOKEN"
}
}
}
}
- Use OpenClaw with the mcporter skill/runtime to list and call tools from
g-gremlin-hubspot.
How it compares
| Capability | HubSpot Official MCP | peakmojo/mcp-hubspot | g-gremlin-hubspot-mcp |
|---|---|---|---|
| CRM search | 10k cap | cached/vector | Auto-window past 10k |
| Write operations | No | Create only | Upsert + dry-run |
| Duplicate detection | No | No | Merge plans |
| Schema introspection | No | No | Full schema/props |
| Engagement export | No | No | Async fallback |
| Snapshot diffing | No | No | Yes |
| Safety layer | N/A | None | Dry-run + plan hash |
| Impact classification | No | No | Per-tool labels |
How we break the 10k ceiling
HubSpot's Search API has a hard cap of 10,000 total results — no cursor beyond that.
g-gremlin breaks this with recursive date-range windowing:
- Initial query hits the 10k ceiling
- Inspects
createdatetimestamps in returned records - Splits the time range into binary halves, re-queries each window
- Repeats recursively (max depth 8, min window 7 days)
- Deduplicates across windows by record key
Required: createdate (default) or any sortable date property. For objects without timestamps, engagements use an async export fallback with no ceiling.
Available tools
7 tools work with no account (FREE). 7 require a g-gremlin HubSpot Admin license or active trial. A 30-day trial is available at first install.
Tier 1: Read & Discover
| Tool | Impact | License | What it does |
|---|---|---|---|
hubspot.auth.whoami |
[READ] |
FREE | Check auth, show portal identity |
hubspot.auth.doctor |
[READ] |
Licensed | Health diagnostics (connectivity, scopes, API access) |
hubspot.schema.list |
[READ] |
FREE | List all CRM object types (standard + custom) |
hubspot.schema.get |
[READ] |
FREE | Full schema for an object type (properties, associations) |
hubspot.props.list |
[READ] |
Licensed | Property introspection (names, types, labels) |
hubspot.objects.query |
[READ] |
FREE | CRM search with filters (Search API, capped at 10k) |
hubspot.objects.pull |
[READ] |
FREE | Full extraction past the 10k ceiling (auto-windowing) |
hubspot.engagements.pull |
[READ] |
FREE | Engagement pull with async export fallback |
Tier 2: Analyze & Plan
| Tool | Impact | License | What it does |
|---|---|---|---|
hubspot.dedupe.plan |
[ANALYZE] |
FREE | Scan for duplicates, generate merge plan with plan_hash |
hubspot.props.drift |
[ANALYZE] |
Licensed | Detect property drift between spec and live portal |
hubspot.snapshot.create |
[READ] |
Licensed | Capture CRM state (schema, props, counts) |
hubspot.snapshot.diff |
[ANALYZE] |
Licensed | Compare two snapshots, show what changed |
Tier 3: Mutate
| Tool | Impact | License | What it does |
|---|---|---|---|
hubspot.objects.upsert |
[WRITE] |
Licensed | Bulk upsert from CSV (dry-run default, two-phase apply) |
hubspot.dedupe.apply |
[MERGE] |
Licensed | Execute a merge plan (requires plan_hash verification) |
The free tools are the strongest hooks: pull past 10k records and generate dedup merge plans — no account needed. The paywall appears when you act on what you found (upsert, apply merges, snapshots).
Safety model
All mutations use two-phase confirmation:
- Dry-run (default): tool runs without making changes, returns a preview +
plan_hash - Apply: caller passes
apply=trueAND theplan_hashfrom step 1. If the hash doesn't match (plan changed, wrong file), the tool rejects with a clear error.
Every tool response includes an impact classification: [READ], [ANALYZE], [WRITE], or [MERGE].
Response envelope
Every tool returns a consistent GremlinMCPResponse/v1 JSON:
{
"$schema": "GremlinMCPResponse/v1",
"ok": true,
"summary": "Pulled 47,231 contacts across 12 auto-window queries",
"data": { ... },
"artifact": { "type": "file", "path": "...", "row_count": 47231 },
"warnings": [],
"safety": { "dry_run": false, "impact": "read" },
"raw": { "agentic_result": { ... }, "exit_code": 0 }
}
Auth
This MCP server never stores tokens. It delegates to g-gremlin's credential chain:
G_GREMLIN_HUBSPOT_ACCESS_TOKENenv var (highest priority)g-gremlin hubspot connect --access-token <PAT>(stored locally in~/.g_gremlin/)g-gremlin hubspot oauth connect(browser-based OAuth)
Troubleshooting
"HubSpot not configured" error in Claude Desktop?
Claude Desktop may run under a different user context than your terminal. Set the token in the MCP config's env block (see Quickstart above) or set G_GREMLIN_HUBSPOT_ACCESS_TOKEN as a system-level environment variable.
Known gaps & roadmap
This is a public beta. Core read and analyze workflows are stable. Known gaps:
- Remote MCP (SSE transport) — currently stdio only; SSE/streamable HTTP planned for teams that don't want local installs
- Workflow diffing — g-gremlin has workflow comparison commands, not yet exposed as MCP tools
- Association management — create/delete associations between objects
- List management — HubSpot list creation and membership management
- Pipeline management — deal/ticket pipeline configuration
Found a bug or have a feature request? Open an issue.
Requires
- Python 3.10+
- g-gremlin >= 0.1.14 (version checked at startup)
- A HubSpot Private App token with CRM scopes
Development
git clone https://github.com/mikeheilmann1024/g-gremlin-hubspot-mcp
cd g-gremlin-hubspot-mcp
pip install -e ".[dev]"
pytest
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。