Parallel Task MCP server
Commit Sep 11, 2026Starts deep research jobs and dataset enrichment tasks and retrieves their results as Markdown without leaving the agent.
Also known as parallel-web/task-mcp, task-mcp
- Stars
- 16
- Tools
- —
- Downloads, 30 days
- —
- Latest version
- 1.0.0
- Last commit
- Sep 11, 2026
- License
- —
What is Parallel Task MCP?
Parallel Task starts deep research jobs and dataset enrichment tasks and retrieves their results as Markdown without leaving the agent.
Summary generated from the project's README
About Parallel Task MCP server
Conducting deep research or enriching datasets directly from your AI assistant is what this server enables. It allows developers to initiate complex information gathering tasks and retrieve the final results as Markdown without leaving their existing workflow. The service is hosted by Parallel, meaning you do not need to run the repository locally to use it, though you can run a local proxy for development purposes.
Users connect via HTTP or Streamable HTTP transport using clients like Claude Code, Codex, Cursor, or VS Code. Setting up the connection requires a Parallel account and authentication through either a browser-based sign-in or an API key. Because creating research or enrichment tasks is a billable activity, users should monitor their Task API pricing and account balance.
The server provides four specific tools: createDeepResearch for detailed questions, createTaskGroup for processing lists of inputs with a common schema, getStatus to monitor active jobs, and getResultMarkdown to fetch the completed output. These tasks run asynchronously, allowing you to check progress while continuing other work.
What you can do with it
- 1Perform deep company researchAsk the agent to write a detailed company brief including products, customers, and funding with cited sources.
- 2Enrich small datasetsProvide a list of companies to extract specific details like headquarters, founding year, and website URLs into a structured format.
- 3Monitor long-running tasksCheck if a research job or a group of enrichment tasks is still active without restarting the process.
- 4Retrieve formatted research resultsGet the final output of a completed research task in Markdown format for easy analysis.
Written from the project's README. Every use case is checked against a quote from it.
How to install Parallel Task MCP
Formats verified 10 Nov 2026Add to CursorAdd to GooseInstall Parallel Task MCP in Claude Code
$ claude mcp add --transport http task-mcp https://task-mcp.parallel.ai/mcpInstall Parallel Task MCP in Cursor
~/.cursor/mcp.json
{
"mcpServers": {
"task-mcp": {
"url": "https://task-mcp.parallel.ai/mcp"
}
}
}Install Parallel Task MCP in VS Code
.vscode/mcp.json
{
"servers": {
"task-mcp": {
"type": "http",
"url": "https://task-mcp.parallel.ai/mcp"
}
}
}Install Parallel Task MCP in VS Code CLI
$ code --add-mcp '{"name":"task-mcp","type":"http","url":"https://task-mcp.parallel.ai/mcp"}'Install Parallel Task MCP in Codex
$ codex mcp add task-mcp --url https://task-mcp.parallel.ai/mcpInstall Parallel Task MCP in Codex config.toml
~/.codex/config.toml
[mcp_servers.task-mcp] url = "https://task-mcp.parallel.ai/mcp"
Install Parallel Task MCP in Gemini CLI
$ gemini mcp add -s user --transport http task-mcp https://task-mcp.parallel.ai/mcpInstall Parallel Task MCP in Gemini settings.json
~/.gemini/settings.json
{
"mcpServers": {
"task-mcp": {
"url": "https://task-mcp.parallel.ai/mcp",
"type": "http"
}
}
}Automated checks
Ran Nov 15, 2026Full report →Not verified yet. A server needs a score of 80, no serious finding, and enough data for us to check. How checks work · What to fix for more points
- Source repository linked and reachableparallel-web/task-mcp is public and reachablegithub.com/parallel-web/task-mcp ↗Pass
- Package names this repository as its sourceNot checked
- Build provenance (Sigstore attestation)Not checked
- Open source license (OSI)No licenseFinding
- Commit or release in the last yearLast commit or release 2026-09-11github.com/parallel-web/task-mcp ↗Pass
- OpenSSF ScorecardNot checked
- Package name not close to a popular packageNot checked
- No critical advisory in the package or its dependenciesNot checked
- No hidden instructions to the model0 tool descriptions and the README read; no override phrases, credential paths or hidden charactersPass
- Destructive tools are labelledNot checked
- Remote endpoint answers the MCP handshakeHTTPS endpoint answered 401 (sign-in required)Pass
README
View on GitHubParallel Task MCP
Deep research and data enrichment from your AI assistant.
Research a question or enrich a small dataset, then retrieve the results without leaving your agent. Parallel hosts the service; you do not need to run this repository to use it.
Task MCP requires a Parallel account and authentication. Unlike the free Search MCP, creating research or enrichment tasks is billable. Review Task API pricing before starting work.
Setup documentation · Get an API key · Task MCP documentation
Manual setup
Choose one connection method for your client. Reuse an existing connection to this endpoint rather than adding it twice.
Claude Code
Run this in the project where you want to use Parallel:
claude mcp add --transport http --scope project parallel-task https://task-mcp.parallel.ai/mcp
Start a new Claude Code session, run /mcp, and complete the Parallel sign-in in your browser when prompted.
Codex
Make your API key available as PARALLEL_API_KEY in the environment that launches Codex, then run:
codex mcp add parallel-task --url https://task-mcp.parallel.ai/mcp \
--bearer-token-env-var PARALLEL_API_KEY
Start a new Codex session and run /mcp to check the connection. The configuration references the environment variable, not the key itself. Do not commit API keys to configuration files or paste them into a conversation. For OAuth instead, see the Codex setup instructions.
Cursor, VS Code, and other clients
Use this server URL with HTTP / Streamable HTTP transport:
https://task-mcp.parallel.ai/mcp
For Cursor, merge this into your MCP configuration:
{
"mcpServers": {
"parallel-task": {
"url": "https://task-mcp.parallel.ai/mcp"
}
}
}
Complete the Parallel sign-in when prompted. Clients without compatible OAuth support can use a Parallel API key as a Bearer token in the Authorization header; use the client's secure credential settings rather than hardcoding the key in a shared file.
VS Code uses a top-level servers object, not mcpServers. Follow the client-specific setup instructions for its configuration format and authentication options.
Available tools
| Tool | What it does |
|---|---|
createDeepResearch |
Starts a research job for a question that needs a detailed, sourced answer. |
createTaskGroup |
Starts enrichment jobs for a list of inputs using a common output schema. |
getStatus |
Checks whether a run or group is still active. |
getResultMarkdown |
Retrieves research or enrichment output as Markdown. |
Verify the connection by checking that these four tools are available. You do not need to start a paid job just to check installation.
Try it
Research a company:
Use
createDeepResearchto research Parallel Web Systems (https://parallel.ai). Write a company brief covering its products, target customers, and publicly announced funding. Cite your sources and flag anything you cannot verify.
Enrich a list of companies:
Use
createTaskGroupto enrich these three companies: Parallel Web Systems (https://parallel.ai), Google (https://google.com), and Apple (https://apple.com). Return one row per company with its official company name, website, headquarters, founding year, and a one-sentence product description. Include source URLs and leave any fields you cannot verify blank.
Research and enrichment run asynchronously:
- Start the job once and keep its run or group ID.
- Use
getStatusto check progress while continuing other work. - Once complete, use
getResultMarkdownto retrieve and analyze the output.
If your client does not continue automatically, ask it to check the existing job in a follow-up message. Do not create another job just to check progress. See the workflow guide for more examples.
Troubleshooting
- Tools are missing: start a new agent session, check the MCP connection list, and confirm the URL is
https://task-mcp.parallel.ai/mcp. - Authentication fails: complete browser sign-in, or check that your client can access a valid API key. For the Codex command above,
PARALLEL_API_KEYmust be present in the environment that launches Codex. - Insufficient credits / HTTP 402: check your balance at platform.parallel.ai before starting another task.
- No results yet: check the existing ID with
getStatus; starting a job does not wait for completion. - Still stuck: see the troubleshooting guide or open an issue. Include your client, version, and error message, but leave out credentials and private task inputs or outputs.
Privacy, support, and security
Task inputs are sent to Parallel to run the requested research or enrichment. Hosted service usage is governed by the Customer Terms and Privacy Policy.
For product support, contact support@parallel.ai.
Running the proxy locally
From a checkout of this repository, with Node.js and npm available:
- Run
npm install. - Run
npx wrangler devand leave it running. - In another terminal, run
npx @modelcontextprotocol/inspector. - Connect the Inspector to
http://localhost:8787/mcpusing Streamable HTTP.
The local proxy still calls the hosted service. Authentication and task charges still apply; this is not an offline task runner.
Parallel Task MCP questions
How do I add Parallel Task MCP to Claude Code?
Run claude mcp add --transport http task-mcp https://task-mcp.parallel.ai/mcp in your terminal. Add --scope user to make it available in every project.
How do I add Parallel Task MCP to Cursor?
Open ~/.cursor/mcp.json and add the server under mcpServers, using the Cursor config shown in the install section above.
What tools does the Parallel Task MCP server provide?
It provides four tools: createDeepResearch, createTaskGroup, getStatus, and getResultMarkdown.
How do I check progress on a research job?
Keep the run or group ID from when you started the job, then call getStatus to check progress while continuing other work.
Do I need a Parallel account to use Task MCP?
Yes, it requires a Parallel account and authentication, and starting research or enrichment jobs is billable.
How do I authenticate the Parallel Task MCP server?
You can complete a browser sign-in via OAuth, or use a Parallel API key as a Bearer token in the Authorization header.