Skip to content
Get packs

Guide · Agent skill

How to use the DeerFlow Agent Bridge agent skill (claude-to-deerflow)

Updated · written with AI from the project's own documentation and checked against our catalogue data

The claude-to-deerflow skill teaches an AI assistant such as Claude how to talk to DeerFlow, a separate multi-agent research platform that you must already be running on your own server or machine. This guide is for readers who have not used agent skills before and want to connect Claude to a DeerFlow instance. By the end you will know what the skill does, how to install it, how to phrase requests so it triggers, and what to check for safety.

DeerFlow Agent Bridgebytedance/deer-flow · MIT

Quick start

  1. 1Download and unzip. Download the skill ZIP from this page and unzip the claude-to-deerflow folder.
  2. 2Place or upload the skill. Put the folder in ~/.claude/skills/ or a project's .claude/skills/, or upload it under Settings → Capabilities → Skills in the Claude apps.
  3. 3Start DeerFlow. Make sure a DeerFlow instance is running and reachable, by default at http://localhost:2026.
  4. 4Set environment variables if needed. If DeerFlow runs elsewhere, set DEERFLOW_URL and optionally DEERFLOW_GATEWAY_URL and DEERFLOW_LANGGRAPH_URL.
  5. 5Ask in plain language. Mention DeerFlow or describe a deep research task and the assistant will use the skill.
  6. 6Review the reply. The skill returns DeerFlow's final AI response; continue on the same thread for follow-ups.

What the claude-to-deerflow skill makes the agent do

An agent skill is a folder of instructions (a SKILL.md file, plus optional scripts) that an assistant reads when a request matches the skill's description. This skill covers the DeerFlow HTTP API. DeerFlow is described as an AI agent platform built on LangGraph that orchestrates sub-agents for research, code execution, web browsing and more.

Once installed and a DeerFlow server is reachable, the assistant can send your message to DeerFlow, start a conversation thread, and return DeerFlow's reply. It can also check whether DeerFlow is online, list its models, skills and agents, upload files to a thread, and view past threads and memory.

  • Send messages or research questions to DeerFlow and collect the response
  • Start a new conversation thread or continue an existing one
  • Check DeerFlow health and status
  • List models, skills and agents; enable or disable a skill
  • Upload PDF, PPTX, XLSX or DOCX files, which DeerFlow converts to Markdown
  • View thread history, list threads and read memory

Install the skill (ZIP download or Claude apps upload)

Download the ZIP of the skill from this page and unzip it. For a personal setup, place the unzipped claude-to-deerflow folder inside ~/.claude/skills/. For a single project, place it inside that project's .claude/skills/ folder instead.

If you use the Claude apps rather than Claude Code, you can upload the skill under Settings → Capabilities → Skills instead of copying files to disk.

  • Personal install: ~/.claude/skills/claude-to-deerflow/
  • Project install: <your-project>/.claude/skills/claude-to-deerflow/
  • Claude apps: Settings → Capabilities → Skills, then upload the skill
~/.claude/skills/claude-to-deerflow/

Make sure a DeerFlow server is running

The skill does not include DeerFlow. It only connects to an instance you already run. By default it expects DeerFlow at http://localhost:2026. The skill's health check calls the gateway's /health endpoint, so you can test the connection yourself first.

If the health check fails, the skill reports that DeerFlow is not running and tells you to start it. Common causes named in the source are a closed port, services that are still starting up, and configuration errors.

  • Default address: http://localhost:2026
  • Health check path: /health on the gateway URL
  • If unreachable, start your DeerFlow instance and wait for services to finish starting
curl -s "http://localhost:2026/health"

Configure the server address with environment variables

If DeerFlow runs on another host, set environment variables before the assistant makes requests. DEERFLOW_URL is the unified proxy base URL. DEERFLOW_GATEWAY_URL covers REST endpoints such as models, skills, memory and uploads, and DEERFLOW_LANGGRAPH_URL covers threads and runs. If you leave the last two unset, they are derived from DEERFLOW_URL.

  • DEERFLOW_URL (default http://localhost:2026)
  • DEERFLOW_GATEWAY_URL (default is the value of DEERFLOW_URL)
  • DEERFLOW_LANGGRAPH_URL (default is DEERFLOW_URL with /api/langgraph appended)
DEERFLOW_URL="${DEERFLOW_URL:-http://localhost:2026}"

Trigger the skill with natural requests

You do not need a special command. The skill's description lists the situations it covers, so ask in plain language. Mentioning DeerFlow or deer flow by name is the clearest signal. Requests for a deep research task that DeerFlow can handle also match.

For research, the source's tips suggest the pro or ultra context mode, which enables planning and sub-agents. For quick questions, flash mode is faster and skips planning.

  • "Ask DeerFlow to compare the main approaches to solid-state batteries and summarise the trade-offs."
  • "Check whether DeerFlow is up and list its available models."
  • "Upload this PDF to a new DeerFlow thread and ask it for the key findings."
  • "Delegate this market research task to DeerFlow and bring back the answer."
  • "Show me my recent DeerFlow threads."

How a message is sent and the reply is read

To send a message, the skill creates a thread and then streams a run into it. The response is a stream of events. The final answer is found in the last values event, in the last message whose type is ai. Reuse the same thread_id to continue a conversation later.

The bundled helper scripts/chat.sh runs the whole sequence: it checks health, creates a thread, streams the run and prints the final AI response.

  • Creates a thread, then streams the run to /threads/<thread_id>/runs/stream
  • Context modes: flash (no thinking, planning or sub-agents), standard, pro (planning) and ultra (planning plus sub-agents)
  • Stream events include metadata, values, messages-tuple and end
bash /path/to/skills/claude-to-deerflow/scripts/chat.sh "Your question here"

Safety notes: scripts and network access

This skill runs scripts and uses the network. The included scripts are chat.sh and status.sh, and they need bash, curl and standard shell tools. Everything the skill does is sent to whatever host DEERFLOW_URL points to. Messages, uploaded documents and thread content leave your machine for that server.

Only point the skill at a DeerFlow instance you run or trust. Do not upload sensitive or private documents to a server you do not control. The skill does no research on its own; without a running DeerFlow it only reports that the server is unreachable.

  • Included scripts: scripts/chat.sh and scripts/status.sh
  • Requires bash, curl and standard shell tools
  • Uploads and messages are sent to the host set in DEERFLOW_URL
  • Read the SKILL.md and scripts before running them on sensitive work

What people use it for

  • Delegate a long multi-step research question to DeerFlow and collect the answer in one place.
  • Upload a set of PDF or DOCX reports to a thread and ask DeerFlow to analyse them.
  • Check whether your DeerFlow server is online and which models and skills it has.
  • Review past conversations by listing threads and reading their history.
  • Script or test DeerFlow's HTTP API from a developer workflow.

Frequently asked questions

Is the DeerFlow Agent Bridge skill free?

The skill is published under the MIT licence, as stated on this page. DeerFlow itself must be running on a server you control, and this page does not state prices for hosting it.

Do I need to install DeerFlow to use this skill?

Yes. The skill only connects to a DeerFlow instance that is already running and reachable over HTTP, by default at http://localhost:2026. Without one, it reports that DeerFlow is unreachable.

Where do I put the skill folder?

Place the unzipped claude-to-deerflow folder in ~/.claude/skills/ for personal use or in a project's .claude/skills/ folder. In the Claude apps, upload it under Settings → Capabilities → Skills.

Why does the health check fail?

The source lists port not open, services still starting up, and config errors as common causes. Confirm DeerFlow is running at the address in DEERFLOW_URL and try the /health endpoint with curl.

How is this different from just asking Claude to do research?

This skill does not do research itself. It sends your message to DeerFlow, which orchestrates sub-agents for research and other tasks, and returns the reply.

Is it safe to upload private documents?

Be careful. Uploads and messages go to whatever host DEERFLOW_URL points to. Only use a server you run or trust, and avoid sending sensitive data to one you do not control.

New agent skills, repos and AI tools — every week

The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.

Our verdict

Use this skill if you already run a DeerFlow instance and want Claude Code or another agent to send it research tasks, uploads and status checks from one place. Skip it if you do not have a DeerFlow server, or if you need an agent to do research on its own or keep sensitive documents off third-party hosts.

Download the skill