A Model Context Protocol (MCP) server for managing Tempo worklogs in Jira. This server provides tools for tracking time and managing worklogs through Tempo's API, making it accessible through Claude, Cursor and other MCP-compatible clients.
MCP server for managing Tempo worklogs in Jira
Estrelas
47
Ferramentas
—
Disponibilidade
—
Features
Retrieve Worklogs: Get all worklogs for a specific date range
Create Worklog: Log time against Jira issues
Bulk Create: Create multiple worklogs in a single operation
Edit Worklog: Modify time spent, dates, and descriptions
Delete Worklog: Remove existing worklogs
Missing Days Report: Find working days where you logged less than expected (uses Tempo's user-schedule, so holidays and non-working days are skipped automatically)
Worklog Analytics: Aggregate hours by issue, account, day, week, or month with totals and percentages
System Requirements
Node.js 18+ (LTS recommended) — only needed for the local stdio modes
Jira Cloud instance
Tempo API token
Jira API token (not required when using OAuth 2.0 PKCE authentication)
Usage Options
There are three ways to use this MCP server:
Remote / Cloudflare Workers (no install) — host once, share with your team. Each user generates their own URL via a setup page and pastes it into Claude.ai or ChatGPT. Works from web and mobile.
NPX: Run directly with npx on your laptop, no clone required.
Local Clone: Clone the repository for development or customization.
If you just want to use the server, option 1 is the easiest and works on phones too. If you're a maintainer deploying for your team, see the Remote deployment guide.
Option 1: Remote (Cloudflare Workers)
For end users
Once your team has the server deployed, the flow is:
Open https://<your-deployment>.workers.dev/setup.
Paste your Tempo API token, Jira base URL, Jira API token, and Jira email. Hit Generate MCP URL.
The page returns a personal URL like https://<your-deployment>.workers.dev/mcp/u_<random>. Copy it.
In Claude.ai → Settings → Connectors → Add custom connector, paste the URL.
The connector syncs across web, desktop, and mobile (iOS/Android).
For ChatGPT: enable Settings → Apps → Advanced → Developer mode (Pro/Plus/Business+), then add the URL as a custom MCP server. Plus/Pro accounts can read; write tools (create/edit worklogs) require Business+ per OpenAI's tier.
The URL contains your credentials — treat it like a password, don't share or commit it.
Remote deployment (Cloudflare Workers)
Free-tier hosting on Cloudflare Workers. ~5–10 minutes from clone to live URL. Anyone can fork and self-host — no upstream coordination needed.
Prerequisites
A Cloudflare account (free plan is enough).
Node.js 18+ and npm locally — only used for wrangler CLI; the Worker runtime itself doesn't run Node.
One-time setup
git clone https://github.com/ivelin-web/tempo-mcp-server.git
cd tempo-mcp-server
npm install
# 1. Log in to Cloudflare (opens browser).
npx wrangler login
# 2. Create your own KV namespace for per-user credentials.
npx wrangler kv namespace create USERS
⚠️ If you forked the repo: the committed wrangler.jsonckv_namespaces[0].id belongs to the upstream maintainer's Cloudflare account. Replace it with the id step 2 just returned, otherwise wrangler deploy will fail with KV namespace … is not valid. KV namespace ids are public per-account identifiers, not secrets, but each account has its own.
# 3. Generate and store the encryption key.
# Used to AES-GCM-encrypt per-user credentials in KV.
openssl rand -base64 48 | npx wrangler secret put ENCRYPTION_KEY
# 4. (Optional) Pin the CORS origin. Defaults to "*". Set it if you only
# want browsers from a specific app to call the Worker.
echo "https://claude.ai" | npx wrangler secret put ALLOWED_ORIGIN
# 5. Deploy.
npm run remote:deploy
# → outputs https://tempo-mcp-server.<your-account>.workers.dev
Visit /setup on the deployed URL to onboard your first user.
Updating an existing deployment
After pulling new commits from upstream:
npm install # picks up any new deps
npm run remote:deploy # ships the new Worker bundle
Secrets and KV data persist across deploys. compatibility_date and compatibility_flags in wrangler.jsonc are pinned, so behaviour doesn't drift silently when Cloudflare ships runtime changes.
Local development
cp .dev.vars.example .dev.vars
# edit .dev.vars and put a real ENCRYPTION_KEY (any value works locally)
npm run remote:dev
# → http://localhost:8787 with a mock KV; data is wiped between sessions
Other useful scripts:
npm run remote:typecheck — type-check the Worker bundle (uses tsconfig.worker.json).
npm run remote:tail — stream live logs from the deployed Worker.
Troubleshooting
KV namespace … is not valid — kv_namespaces[0].id in wrangler.jsonc is empty (or wrong). Run npx wrangler kv namespace create USERS and paste the new id.
ENCRYPTION_KEY is not defined at runtime — secret wasn't set. Re-run step 3.
Rate limit binding … not available — your account's plan doesn't include the Workers Rate Limiting API. Either upgrade, or remove the ratelimits block in wrangler.jsonc and the SETUP_RATE_LIMITER.limit(...) call in src/remote/worker.ts.
404 from /mcp/u_… — the user id is unknown (or never existed). The Worker returns 404 by design for invalid/missing ids; have the user re-run /setup.
Existing users suddenly can't connect — most likely cause is a rotated ENCRYPTION_KEY; existing AES-GCM blobs can't be decrypted with the new key. See the warning below.
How it works
Credential storage: the /setup POST handler AES-GCM encrypts the form data with ENCRYPTION_KEY and stores it in KV under user:u_<random>. Each MCP request reads + decrypts that record, builds an McpServer for that single request, and dispatches via Cloudflare's official createMcpHandler. No credentials are held in memory between requests; no Durable Objects are used.
Treat ENCRYPTION_KEY as long-lived. Rotating it invalidates every existing user record (the AES-GCM tag won't validate against the new key), and all your users will need to re-run /setup. Pick a key from openssl rand -base64 48 once and never change it.
Auth model: the URL /mcp/u_<id>is the credential. The 22-char base64url id carries ~128 bits of entropy. We never return 401 for that path (Claude.ai web has known bugs around the 401-then-OAuth flow), and we return 404 for unknown ids. This matches the URL-token pattern used by Zapier MCP, Pipedream MCP, and similar.
Hardening already in place:
Per-IP rate limit (5 req/min) on POST /setup, via Cloudflare's native Rate Limiting binding.
Cache-Control: no-store on /setup responses so the success page (which contains the MCP URL) and the error re-render (which echoes tokens back) never sit in any cache.
Referrer-Policy: no-referrer on every HTML page so the MCP URL doesn't leak via referrer headers.
Limits to know about:
Workers free plan: 100k requests/day, 50ms CPU per request (we are I/O-bound, comfortable).
KV free plan: 100k reads/day, 1k writes/day. Setup writes once per user; reads happen per MCP call.
The Worker only supports Jira basic auth (classic API token + email). Bearer and the OAuth 2.0 PKCE flow are stdio-only — bearer requires gateway URL routing that the Worker does not yet do, and PKCE needs a browser callback the Worker can't host.
Option 2: NPX Usage
The easiest way to use this server is via npx without installation:
Connecting to Claude Desktop (NPX Method)
Open your MCP client configuration file:
Claude Desktop (macOS): ~/Library/Application Support/Claude/claude_desktop_config.json
Claude Desktop (Windows): %APPDATA%\Claude\claude_desktop_config.json
Click "Create API token" (the classic, unscoped flow). This is what works with basic auth out of the box.
Do not use "Create API token with scopes" — those tokens must be sent through Atlassian's gateway URL (https://api.atlassian.com/ex/jira/{cloudId}/...) with the cloud ID, which this server's basic auth path does not currently route to. They will fail with 401 against your site URL. If you only have a scoped token available (e.g. your org disabled classic tokens), use the OAuth 2.0 PKCE flow instead — it routes through the gateway automatically.
Environment Variables
The server requires the following environment variables:
TEMPO_API_TOKEN # Your Tempo API token
JIRA_API_TOKEN # Your Jira API token (required for basic and bearer auth)
JIRA_EMAIL # Your Jira account email (required for basic auth)
JIRA_BASE_URL # Your Jira instance URL (e.g., https://your-org.atlassian.net)
JIRA_AUTH_TYPE # Optional: 'basic' (default), 'bearer', or 'oauth'
JIRA_OAUTH_CLIENT_ID # OAuth 2.0 client ID (required for oauth auth)
JIRA_OAUTH_CLIENT_SECRET # OAuth 2.0 client secret (required for oauth auth)
JIRA_TEMPO_ACCOUNT_CUSTOM_FIELD_ID # Optional: Custom field ID for Tempo accounts
You can set these in your environment or provide them in the MCP client configuration.
Authentication Types
The server supports three authentication methods for the Jira API:
Basic Authentication (default)
Uses email and API token. This is the traditional method:
Note: When using bearer auth, JIRA_EMAIL is not required as the user is identified from the token.
OAuth 2.0 PKCE Authentication
Some Atlassian organizations restrict API token access via admin policy, which causes basic and bearer authentication to fail. The oauth type implements the full OAuth 2.0 authorization code flow with PKCE and works regardless of API token restrictions — tokens are short-lived and refreshed automatically without any manual management.
On first use, a browser window opens for you to authorize access. Tokens are stored locally at ~/.tempo-mcp-server/tokens.json and refreshed automatically.
Create an OAuth 2.0 app in Atlassian Developer Console with the read:jira-user and read:jira-work scopes and http://localhost:7788/callback as the callback URL.
Note: JIRA_API_TOKEN and JIRA_EMAIL are not required when using oauth auth.
Tempo Account Configuration
If your Tempo instance requires worklogs to be linked to accounts, set the custom field ID that contains the account information:
JIRA_TEMPO_ACCOUNT_CUSTOM_FIELD_ID=10234
To find your custom field ID:
Go to Jira Settings → Issues → Custom Fields
Find your Tempo account field and note the ID from the URL or field configuration
Viewing Other Users' Worklogs
The read tools (retrieveWorklogs, getWorklogAnalytics, getMissingWorklogDays) default to the token owner's own worklogs, but accept optional filters to target other people — useful for admins, PMs, and team leads:
users — array of emails, display names, or Jira accountIds (e.g. ["[email protected]", "Maria Petrova"])
program — Tempo Program name or id; expands to all current members of the program's teams
team — Tempo Team name or id; expands to all current team members
Filters combine as a union. Tempo enforces permissions server-side: the API silently omits worklogs the token owner isn't allowed to see (no error is returned). To view other users you need:
A Tempo Permission Role with View Worklogs granted to the token owner (Tempo > Settings > Permission Roles) — "Full" for everyone, or "Restricted" + selected teams
Jira Browse Projects permission on the relevant projects
The Teams scope on the Tempo API token if you use the program / team filters
getWorklogAnalytics also supports groupBy: "user" — combined with program, it produces a per-person hours report in a single call. getMissingWorklogDays with a filter returns a per-user report of days with missing time (viewing other users' schedules is also permission-gated).
Available Tools
retrieveWorklogs
Fetches worklogs for the configured user (or other users via filters) between start and end dates.
Parameters:
- startDate: String (YYYY-MM-DD)
- endDate: String (YYYY-MM-DD)
- users: String[] (optional) — emails, display names, or accountIds
- program: String (optional) — Tempo Program name or id
- team: String (optional) — Tempo Team name or id
Reports working days in a date range where the user has logged less time than expected. Expected hours per day come from the user's Tempo schedule, so holidays, non-working days, and part-time schedules are honoured automatically. With users / program / team it checks other people and returns a per-user report (sorted by most missing hours).
Parameters:
- startDate: String (YYYY-MM-DD)
- endDate: String (YYYY-MM-DD)
- minHoursPerDay: Number (optional) — override the per-day threshold;
non-working days are still skipped
- users: String[] (optional) — emails, display names, or accountIds
- program: String (optional) — Tempo Program name or id
- team: String (optional) — Tempo Team name or id
Required Tempo scope: the TEMPO_API_TOKEN must include the Schemes scope (covers Workload Schemes, Holiday Schemes, User Schedule) in addition to Worklogs. Tempo does not allow modifying scopes on an existing token — if your current token only has Worklogs, create a new one at Tempo > Settings > API Integration.
getWorklogAnalytics
Aggregates worklogs in a date range and returns hours, worklog count, and percentage per group, sorted by hours descending. Combine groupBy: "user" with program / team / users for a per-person report.
Parameters:
- startDate: String (YYYY-MM-DD)
- endDate: String (YYYY-MM-DD)
- groupBy: "issue" | "account" | "user" | "day" | "week" | "month" (optional, default "issue")
- users: String[] (optional) — emails, display names, or accountIds
- program: String (optional) — Tempo Program name or id
- team: String (optional) — Tempo Team name or id
Ainda não há lista de ferramentas. Servidores remotos recebem uma após a primeira verificação de saúde.
Perguntas frequentes
Perguntas
Como instalo Tempo?
Copie nesta página a configuração stdio para Cursor, Claude Desktop, Claude Code ou VS Code. Ela usa a instalação do publicador, não um proxy do PlugMCP.
Tempo precisa de uma chave de API?
Este registro não lista nenhuma variável de ambiente obrigatória.
Qual transporte Tempo usa?
Tempo roda localmente via stdio.
Quando Tempo foi verificado pela última vez?
Última verificação: 2026-10-28. MCP server for managing Tempo worklogs in Jira