asana-connector

asana-connector

Enables interaction with Asana projects and tasks via the Asana API, supporting listing projects/tasks, creating/updating tasks, and adding comments, with secure authentication and rate-limit handling.

Category
Visit Server

README

Asana Connector

Project management integration console · DOO Builders League

Builder: Idrees Khaled · Provider: Asana · Version: v1.0.0

A production-oriented Asana connector: a reusable connector core with five typed actions, normalized errors, pagination and rate-limit handling, a thin MCP adapter, and a dark-first developer console over the top.


Quickstart

Runs with no credentials and no configuration. A fresh clone boots into clearly-labelled demo mode, so you can review the whole surface immediately.

git clone https://github.com/Idrees-02/doo-asana-connector.git
cd doo-asana-connector
npm install
npm run dev

Requires Node ≥ 20.19 and nothing else. No Docker, no global installs, no hosted services.

Reviewing the engineering rather than the UI

npm test              # connector suite: all five actions, errors, pagination, rate limits
npm run test:frontend # console tests
npm run verify        # typecheck + lint + secret scan + tests
npm run mcp           # MCP server over stdio

The five actions

These IDs are fixed by the assignment and are used verbatim throughout the manifest, schemas, OpenAPI spec, MCP tools and console.

Action Type Asana endpoint
asana.list_projects READ GET /projects
asana.list_project_tasks READ GET /projects/{gid}/tasks
asana.create_task WRITE POST /tasks
asana.update_task WRITE PUT /tasks/{gid}
asana.add_comment WRITE POST /tasks/{gid}/stories

There is deliberately no delete action — it is not part of the assignment, and the connector never requests a :delete OAuth scope.


Getting your Asana credentials

Never paste a token into a chat window, an issue, a screenshot, or any file that git tracks. It goes into .env on your machine and nowhere else. .env is gitignored, and a pre-commit hook blocks it even if that is bypassed.

Option A — Personal Access Token (recommended to start)

A PAT is the quickest path and is all you need for local review.

  1. Sign in at https://app.asana.com — a free account is enough.

  2. Open https://app.asana.com/0/my-apps (or: profile photo → My Settings → Apps → Manage Developer Apps).

  3. Click Create new token.

  4. Name it something identifiable, e.g. doo-asana-connector-dev.

  5. Accept the API terms, then click Create token.

  6. Copy it immediately — Asana shows it exactly once.

  7. Run the setup command and paste it when prompted:

    npm run setup
    

    Input is hidden while you type, so the token never appears on screen, in a screen recording, or in your shell history. It is written to .env with permissions 0600 and verified against Asana immediately — without ever being displayed back to you.

    Prefer doing it by hand? cp .env.example .env, then edit the ASANA_ACCESS_TOKEN= line in your editor.

  8. Restart npm run dev. The amber DEMO MODE banner disappears and Settings → Test Connection shows your real account and workspaces.

.env.example is tracked by git and is public. Only ever put placeholders there. Real values belong in .env, which is gitignored.

Tip: actions performed through a PAT are attributed to you in Asana's activity feed. For a shared or demo setup, create a dedicated bot user and generate the PAT from that account instead.

Option B — OAuth 2.0 (multi-user)

Needed only if you want the browser-based connect flow rather than a pre-shared token.

  1. Same page: https://app.asana.com/0/my-apps → Create new app.

  2. Name the app and accept the terms.

  3. Under OAuth, add this exact redirect URL:

    http://localhost:8787/api/auth/oauth/callback
    
  4. Copy the Client ID and Client Secret into .env:

    ASANA_OAUTH_CLIENT_ID=...
    ASANA_OAUTH_CLIENT_SECRET=...
    ASANA_OAUTH_REDIRECT_URI=http://localhost:8787/api/auth/oauth/callback
    
  5. Restart, then use Settings → Connect with Asana.

Scopes requested (least privilege — nothing more than the five actions need):

projects:read  tasks:read  tasks:write  stories:write  users:read  workspaces:read

Recommended: a sandbox workspace

Before running any write action, create a throwaway project (e.g. Connector Sandbox) and point the connector at it. Write tests then never touch anything that matters. The connector never deletes anything, but it does create tasks and comments.


Secrets policy

Every credential lives in .env and nowhere else. This is enforced mechanically, not by convention:

  • src/config.ts is the only module permitted to read process.env. An ESLint rule (no-restricted-properties) fails the build if anything else touches it, so credentials cannot spread through the codebase.
  • describeConfig() is the only way config reaches a log, an API response or the UI, and its return type has no field capable of carrying a secret value — presence booleans and an opaque fingerprint only.
  • npm run secrets:scan pattern-scans tracked files (Asana PAT format, bearer tokens, private keys, credential-shaped assignments) and runs in CI.
  • A pre-commit hook blocks staged .env files outright and re-scans staged content.
  • The test suite requires no credentials, so CI runs with no secrets configured at all.

Configuration

All configuration is environment-driven (12-factor), so the same build runs locally and deployed with nothing changed but the environment. See .env.example for every variable with inline documentation.

Variable Default Purpose
ASANA_MODE auto auto | live | demo. auto picks live when credentials exist
ASANA_ACCESS_TOKEN — Personal Access Token
ASANA_RATE_LIMIT_RPM 140 Client-side throttle, just under the 150/min free tier
ASANA_TIMEOUT_MS 15000 Per-request timeout
ASANA_MAX_CONCURRENCY 8 In-flight request cap (Asana allows 50 GET / 15 write)
PORT 8787 API port
MCP_TRANSPORT stdio stdio locally, http for a deployed endpoint

ASANA_MODE=live without credentials fails at startup on purpose — silently serving synthetic data to someone who asked for real data would be the worst possible failure mode.


Project structure

doo-asana-connector/
├── connector.yaml          # generated manifest
├── openapi.yaml            # generated from the same Zod schemas
├── src/
│   ├── connector.ts        # DooConnector: manifest, testConnection, listActions, execute
│   ├── client.ts           # Asana HTTP client: pagination, throttle, retry classification
│   ├── config.ts           # the only reader of process.env
│   ├── auth/               # PAT + OAuth 2.0
│   ├── actions/            # the five actions
│   ├── schemas/            # Zod schemas — single source of truth
│   ├── errors/             # normalized error system
│   ├── runtime/            # shared execution pipeline
│   └── demo/               # demo provider
├── mcp/server.ts           # thin MCP adapter
├── server/                 # HTTP API consumed by the console
├── frontend/               # the console
├── tests/ · fixtures/ · examples/ · docs/
└── .env.example

Documentation

Document Contents
docs/AUTHENTICATION.md Getting a PAT or OAuth app, and how credentials are handled
docs/WRITE-SAFETY.md Why writes are never auto-retried, approval, idempotency, concurrency
docs/LIMITATIONS.md What is not built, and what is not yet verified
openapi.yaml Generated API contract
connector.yaml Generated connector manifest

In-app documentation is also available at /docs in the running console.


MCP

npm run mcp           # stdio — Claude Desktop, MCP Inspector
npm run mcp:inspect   # interactive tool explorer

Claude Desktop (claude_desktop_config.json), replacing the path with this project's absolute location:

{
  "mcpServers": {
    "asana-connector": {
      "command": "npx",
      "args": ["tsx", "/ABSOLUTE/PATH/TO/doo-asana-connector/mcp/server.ts"]
    }
  }
}

The adapter iterates connector.listActions() and registers each as a tool. It contains no Asana endpoint, no schema and no business logic — a test asserts the exposed tool ids equal the connector's action ids, so it cannot drift.


Assignment checklist

Marked honestly. Anything not demonstrated is called out rather than assumed.

Requirement Status
Manifest exists Yes — generated, connector.yaml, 35 actions
Asana authentication (PAT + OAuth 2.0) PAT verified live; OAuth authorize step verified live against Asana (real client_id, PKCE, scopes accepted); interactive consent not clicked through — see below
testConnection has no side effects Yes — asserted by test (no non-GET request)
All five required actions implemented Yes — end-to-end tested, plus 30 extended actions (35 total)
Typed input/output schemas Yes — Zod, single source of truth
Inputs validated Yes — before any network call
Errors normalized Yes — 19 ASANA_* codes
Request IDs Yes — connector-generated; Asana returns none
Retry classification Yes — including manual_with_idempotency_key
Pagination Yes — cursor-based, all list actions
Rate limits handled Yes — client-side pacing before sending
Approval / idempotency / duplicates documented Yes — docs/WRITE-SAFETY.md, all 21 write actions require approval
No secrets committed Yes — scanner + pre-commit hook + CI
Unit and fixture tests pass Yes — 263 total (238 connector + 25 console)
OpenAPI exists Yes — generated, 35 endpoints
MCP adapter exists, duplicates no logic Yes — enforced by test, all 35 actions exposed as tools
Frontend connected to the real backend Yes — no mocked UI data, all 35 actions surfaced
Frontend responsive and accessible Yes — per-breakpoint layouts, 25 tests
Documentation and known limitations Yes
Versioned v1.0.0 Yes
Real sandbox/test-account flow Verified — required 5 actions live end-to-end (9/9), 30 extended actions live-checked via asana.get_current_user
OAuth interactive consent Not clicked through — requires a human login, which this assistant will not perform. Authorization request verified live; token exchange/refresh/revoke covered by 21 tests against a double matching Asana's real contract
HTTPS MCP endpoint deployed Not met by design — local-first; HTTP transport implemented but undeployed

The remaining gap is the HTTPS MCP endpoint, which is not met by design (local-first). See docs/LIMITATIONS.md.


Scripts

Command Purpose
npm run dev API + console together
npm test Connector suite (141 tests)
npm run test:frontend Console suite (25 tests)
npm run verify typecheck + lint + secret scan + tests
npm run generate Regenerate openapi.yaml and connector.yaml
npm run setup Interactive .env setup — hidden token input, verifies the connection
npm run smoke:live Read-only check against real Asana (needs a PAT)
npm run smoke:live -- --writes Also exercises create/update/comment
npx tsx examples/use-connector.ts Use the connector as a library

License

MIT

Recommended Servers

playwright-mcp

playwright-mcp

A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.

Official
Featured
TypeScript
Magic Component Platform (MCP)

Magic Component Platform (MCP)

An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.

Official
Featured
Local
TypeScript
Audiense Insights MCP Server

Audiense Insights MCP Server

Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.

Official
Featured
Local
TypeScript
VeyraX MCP

VeyraX MCP

Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.

Official
Featured
Local
graphlit-mcp-server

graphlit-mcp-server

The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.

Official
Featured
TypeScript
Kagi MCP Server

Kagi MCP Server

An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.

Official
Featured
Python
Neon Database

Neon Database

MCP server for interacting with Neon Management API and databases

Official
Featured
E2B

E2B

Using MCP to run code via e2b.

Official
Featured
Exa Search

Exa Search

A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.

Official
Featured
Qdrant Server

Qdrant Server

This repository is an example of how to create a MCP server for Qdrant, a vector search engine.

Official
Featured