Gealo MCP Server
Connects AI clients like Claude, Cursor, and VS Code to a Gealo workspace's tasks, projects, meetings, documents, and chat with OAuth authentication and tenant isolation.
README
Gealo MCP Server
Hosted remote Model Context Protocol server for Gealo workspaces.
Connect Claude, Cursor, VS Code, Codex, or Gemini CLI to your organization's tasks, projects, meetings, documents, and chat — with OAuth 2.1, tenant isolation, and the same project permissions you have in the app.
The MCP server runs on Gealo's infrastructure at your tenant URL (
https://{slug}.gealo.app/mcp). This repository documents the public integration surface and ships an optional STDIO bridge Docker image for MCP clients that only support local subprocess transport.
Endpoint
Every Gealo organization has its own MCP endpoint:
https://{your-tenant-slug}.gealo.app/mcp
Example: workspace acme → https://acme.gealo.app/mcp
Copy the exact URL from Tenant → AI Agents after sign-in.
Discovery
OAuth protected-resource metadata is published per tenant:
GET https://{your-tenant-slug}.gealo.app/.well-known/oauth-protected-resource/mcp
Transport
Gealo's hosted MCP server uses Streamable HTTP at /mcp (MCP 2025-03-26+). It also supports session-backed SSE when clients send Accept: text/event-stream and maintain an Mcp-Session-Id. stdio is not served by Gealo — clients that only speak stdio need a local bridge (see Docker image below).
| Transport | Gealo hosted server | This Docker image |
|---|---|---|
| Streamable HTTP | Yes (primary) | Proxied to remote |
| SSE (session) | Yes (optional) | Proxied when client requests it |
| stdio | No | Local side only (bridge speaks stdio to your MCP client) |
Authentication
- Interactive (default): OAuth 2.1 + PKCE. Your MCP client (or the bridge) opens a browser consent flow on first connect. Tokens are stored locally by the client/bridge (
~/.mcp-authorMCP_REMOTE_CONFIG_DIR), not on Gealo beyond server-side refresh-token rotation metadata. - Scopes:
mcp:read(default),mcp:write(when tenant admin enables write tools) - Headless / CI: Service accounts via
client_credentialsgrant (Tenant → AI Agents → Service Accounts)
curl -X POST "https://{slug}.gealo.app/oauth/mcp/token" \
-d grant_type=client_credentials \
-d client_id=sa:{service-account-id} \
-d client_secret={api-key} \
-d scope=mcp:read \
-d resource=https://{slug}.gealo.app/mcp
Use the returned access_token as Authorization: Bearer on POST /mcp.
Quick connect (recommended — no Docker)
Replace {slug} with your tenant slug. Prefer direct remote URL when your client supports HTTP + OAuth natively.
Cursor
~/.cursor/mcp.json or project .cursor/mcp.json:
{
"mcpServers": {
"gealo": {
"url": "https://{slug}.gealo.app/mcp"
}
}
}
Cursor opens the OAuth consent flow on first use.
See also: examples/cursor-mcp.json
Claude (Desktop / Web)
- Settings → Connectors → Add custom connector
- URL:
https://{slug}.gealo.app/mcp - Advanced → OAuth Client ID:
claude(secret empty) - Complete sign-in and consent in the browser
VS Code (Copilot agent mode)
.vscode/mcp.json:
{
"servers": {
"gealo": {
"type": "http",
"url": "https://{slug}.gealo.app/mcp"
}
}
}
See also: examples/vscode-mcp.json
Codex CLI
~/.codex/config.toml:
[mcp_servers.gealo]
url = "https://{slug}.gealo.app/mcp"
Then: codex mcp login gealo
Gemini CLI
~/.gemini/settings.json:
{
"mcpServers": {
"gealo": {
"httpUrl": "https://{slug}.gealo.app/mcp"
}
}
}
Then /mcp auth gealo if needed.
Docker image (optional STDIO bridge)
Image: ghcr.io/mhdyousuf/gealo-mcp:latest
Purpose: Lets stdio-only MCP clients connect to Gealo's hosted Streamable HTTP endpoint. The container is not the Gealo MCP server — it is a local proxy.
Use Docker when:
- Your MCP client only supports
command/ stdio servers - You need OAuth bridging for a client without native remote MCP support
Do not use Docker when your client supports url / HTTP remote MCP (Cursor, VS Code, Claude Web, Codex, Gemini) — connect directly to https://{slug}.gealo.app/mcp.
Environment variables
| Variable | Required | Default | Description |
|---|---|---|---|
GEALO_TENANT_SLUG |
Yes* | — | Tenant slug (e.g. acme → https://acme.gealo.app/mcp) |
GEALO_MCP_URL |
Yes* | — | Full MCP URL override (alternative to slug) |
GEALO_AUTH_MODE |
No | oauth |
oauth (interactive PKCE) or service_account (headless) |
GEALO_TRANSPORT |
No | http-only |
Remote transport strategy passed to the bridge (http-only recommended) |
GEALO_SCOPE |
No | mcp:read |
OAuth / service-account scope |
GEALO_CLIENT_ID |
SA only | — | Service account ID (sa:{uuid}) |
GEALO_CLIENT_SECRET |
SA only | — | Service account API key |
GEALO_OAUTH_CALLBACK_PORT |
No | auto | OAuth callback port (map with -p in Docker) |
GEALO_ENABLE_HTTP_PROXY |
No | false |
Set true to honor HTTP_PROXY / HTTPS_PROXY |
MCP_REMOTE_CONFIG_DIR |
No | /home/gealo/.mcp-auth |
OAuth token cache directory (mount a volume) |
* Set GEALO_TENANT_SLUG or GEALO_MCP_URL.
Security: Never bake secrets into the image. Pass service-account credentials via env at runtime. Mount a named volume for OAuth tokens so reconnects do not re-prompt.
Local testing
# Build
docker build -t ghcr.io/mhdyousuf/gealo-mcp:local .
# Health check (discovery endpoint reachable)
docker run --rm -e GEALO_TENANT_SLUG=acme ghcr.io/mhdyousuf/gealo-mcp:local dist/health.js
# Interactive OAuth bridge (stdio — pipe JSON-RPC for manual tests)
docker run -i --rm \
-e GEALO_TENANT_SLUG=acme \
-p 3333:3333 \
-v gealo-mcp-auth:/home/gealo/.mcp-auth \
ghcr.io/mhdyousuf/gealo-mcp:local
# Service account (headless)
docker run -i --rm \
-e GEALO_TENANT_SLUG=acme \
-e GEALO_AUTH_MODE=service_account \
-e GEALO_CLIENT_ID=sa:YOUR-SA-ID \
-e GEALO_CLIENT_SECRET=YOUR-API-KEY \
ghcr.io/mhdyousuf/gealo-mcp:local
# Node (without Docker)
npm install && npm run build
GEALO_TENANT_SLUG=acme npm start
Docker + Cursor (stdio clients)
See examples/cursor-mcp-docker.json. Map port 3333 for the OAuth browser callback on first connect.
Docker + VS Code
See examples/vscode-mcp-docker.json.
Docker + service accounts / CI
See examples/cursor-mcp-service-account-docker.json.
Publishing
GitHub Actions workflow .github/workflows/publish-docker.yml publishes to GHCR using GITHUB_TOKEN:
ghcr.io/mhdyousuf/gealo-mcp:latest(onmain)ghcr.io/mhdyousuf/gealo-mcp:<version>(frompackage.jsonand git tagsv*)
Make the package public in GitHub → Packages after the first successful publish.
First tool call
After connecting, call whoami to ground the session:
- User, tenant, and plan
- Remaining monthly MCP tool budget
- Whether write tools are enabled for your token
Use search_tools to browse the catalog filtered to your permissions.
Tools (60+)
Tools are scoped to your tenant and project permissions. Writes require admin opt-in (mcp:write scope).
| Group | Tools |
|---|---|
| Core | whoami, search_tools, execute_tool, apply_pending_action |
| Projects | list_projects, get_project_overview, archive_project, list_task_statuses, create_task_status, update_task_status, delete_task_status, reorder_task_statuses, list_custom_fields, create_custom_field, update_custom_field, delete_custom_field |
| Tasks | search_tasks, get_task, create_task, update_task, assign_task, bulk_update_task_status, add_comment, list_task_comments, add_task_comment, update_task_comment, delete_task_comment, list_task_dependencies, add_task_dependency, remove_task_dependency |
| Sprints & releases | list_sprints, get_sprint, create_sprint, update_sprint, start_sprint, complete_sprint, delete_sprint, list_releases, get_release, create_release, update_release, delete_release, set_task_releases |
| Meetings & chat | list_meetings, get_meeting, search_messages |
| Documents & search | search_documents, semantic_search, answer_from_workspace, find_related_docs |
| Insights | summarize_project, summarize_workspace, find_blockers, review_sprint, compare_releases, summarize_meeting, extract_action_items, get_recent_activity |
| Workforce | my_work, my_leave_balance, workspace_health, usage_report |
| Notifications | list_notifications, get_unread_notification_count, mark_notification_read, mark_all_notifications_read |
Catalog tools (e.g. sprint/release write ops) are reachable via search_tools → execute_tool when your role allows.
Requirements
- A Gealo workspace with MCP access on your plan
- Active membership in that tenant (agents act as you)
- MCP not disabled by a workspace admin
Governance
- Tenant isolation — every call is scoped to your workspace
tenant_id - Project permissions — same templates as the REST API; no privilege escalation via MCP
- Read by default — writes off until an owner enables them
- Destructive preview — archive, bulk status, delete comment, etc. return a preview token;
apply_pending_actionconfirms (10-minute expiry) - Audit & metering — tool calls logged; monthly caps from your subscription entitlements
Details: gealo.app/mcp-server/governance
Multi-tenant users
Connections are per organization. Add one server entry per tenant slug you work in.
Troubleshooting
| Symptom | Likely cause |
|---|---|
401 |
Token expired or revoked — reconnect |
policy_violation:mcp_disabled |
Admin disabled MCP for the tenant |
policy_violation:mcp_client_blocked |
Client blocked in Tenant → AI Agents |
limit_reached:mcpAccess |
Plan does not include MCP |
limit_reached:mcpMonthlyToolCalls |
Monthly tool budget exhausted |
policy_violation:mcp_write_disabled |
Write tools not enabled — ask admin |
forbidden on writes |
Token lacks mcp:write — re-authorize |
| Few or no tools | Your role lacks module permissions |
Links
- Product: gealo.app
- MCP overview: gealo.app/mcp-server
- Governance: gealo.app/mcp-server/governance
- Sign up: gealo.app/register
License
MIT — see LICENSE. Gealo is a commercial product; this repo documents the public MCP integration surface only.
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。