OPENPULSE / FOR AGENTS

Discover open source.
Without leaving your agent.

Find promising projects, compare real growth and inspect AI tool attribution. Ask in plain language; get dated evidence and source links.

Claude Code · Codex · HermesNo API key requiredRead-only data
01 / GET CONNECTED

One message. Your agent does the setup.

MCP + Skill

Copy this message into your agent. It will read the setup guide, configure the connection and verify a real query.

Install OpenPulse using https://openpulse.info/install.md, then find 5 promising open-source projects with source links.

Requires an agent that can configure MCP. A new session may be needed to load tools.Read the setup guide ↗
02 / SEE THE RESULT

One question. A shortlist with substance.

Real projects, observed growth and source links. Here is what your agent can work with.

YOU ASK

Find 3 Rust projects under 5,000 stars with positive seven-day growth. Include activity dates and source links.

RustUnder 5k stars7-day growth
Uses existing observations. No collection or analysis is started.
Finding a real shortlist…Reading existing OpenPulse observations.
03 / MAKE IT YOURS

Three ways to put it to work.

Copy a starting point. Change the language, projects or topic to fit your work.

01 / DISCOVER

Find projects before they get big

Explore smaller projects with observed growth. Get concrete numbers and a reason to look closer.

Find 5 Python projects under 5,000 stars with positive seven-day growth. Explain each choice using observed data and source links.

A shortlist with evidence
02 / COMPARE

Choose with the evidence in view

Compare growth, development activity, licenses and AI index. Keep dates and missing data visible.

Compare psf/requests, encode/httpx and aio-libs/aiohttp using OpenPulse. Show growth, activity periods, licenses and AI index in a table. Mark unavailable data and link each source.

A comparison you can inspect
03 / FOLLOW

Keep an eye on your corner

Follow newly discovered projects in your language or topic. Discovery dates stay separate from creation dates.

Show 5 newly discovered Rust projects with first-seen dates and source links. Explain which growth and activity observations are available.

A dated discovery digest
04 / KEEP UP AUTOMATICALLY

A daily briefing, on your schedule.

Ask Hermes or your existing scheduler to fetch a shortlist, check for changes and save a concise digest.

The scheduler or gateway must stay running. Installing MCP does not create a schedule.
Daily digest example

Help me set up a daily OpenPulse digest of newly discovered Python projects. Confirm the time, timezone and local output path first. Use repository IDs and observation dates to skip processed results, and update progress only after saving successfully. Use my agent’s scheduler if available; otherwise explain how to schedule it.

Manual setup & details

Prefer a terminal command? Everything is here.

Manual MCP setup · Claude Code, Codex & Hermes

Claude Code

claude mcp add --transport http --scope user openpulse https://openpulse.info/mcp

Codex

codex mcp add openpulse --url https://openpulse.info/mcp

Hermes Agent

hermes mcp add openpulse --url https://openpulse.info/mcp
More prompts · AI index & Weekly

Inspect AI tool attribution

Explicit signals in sampled commits, with sample size and evidence.

Show the AI index for Zagitta/gpu-unlock, including the sample size, tool shares and supporting evidence.

Catch up on the shortlist

Read the latest published report with its original dated evidence.

Summarize the latest OpenPulse weekly report. Give one concrete reason to explore each project and link the report.
CLI · Terminal & scripts

Install the CLI

Node.js 20 or later. A downloadable package with no runtime dependencies.

npm install -g https://openpulse.info/downloads/openpulse-cli-1.0.0.tgz
openpulse trending --language Rust --limit 5 --json
openpulse weekly latest --markdown

Use openpulse --help for filters and history queries.

Uninstall: npm uninstall -g @openpulse/cli

Read the CLI source ↗
Skill · Better answers & citations

Add the OpenPulse skill

The skill teaches your agent how to interpret metrics, cite evidence and prepare scheduled digests. It does not configure MCP or create a schedule.

# Codex
mkdir -p ~/.agents/skills/openpulse
curl -fsSL https://openpulse.info/downloads/openpulse/SKILL.md \
  -o ~/.agents/skills/openpulse/SKILL.md

# Claude Code: ~/.claude/skills/openpulse/SKILL.md
# Hermes: ~/.hermes/skills/openpulse/SKILL.md
Read the skill before installing ↗
Advanced scheduling · Hermes & system cron

Hermes daily briefing

After connecting OpenPulse, ask Hermes:

Every day at 09:00 Asia/Shanghai, use OpenPulse to find 5 newly discovered Python projects. Include dated growth and development evidence and source links. Save the digest locally. Keep processed project IDs and observation timestamps in ~/.hermes/openpulse-python-progress.json; update it only after saving successfully. If nothing changed, skip the digest.

For message delivery, choose a destination already connected to your agent. The scheduler or gateway must remain running.

Schedule without an agent

An example for an existing system cron. Times use the scheduler's configured timezone. Use absolute executable and output paths.

0 9 * * * /absolute/path/to/openpulse trending --language Python --limit 5 --json --output /absolute/path/to/openpulse-daily.json

For a weekly agent digest, compare data.report_id with your last successfully processed report. Only advance progress after successful processing. Add backoff for 429 or temporary failures.

Tools & data notes

search_repositoriesget_rankingsget_repositoryget_growth_historyget_ai_indexget_weekly_report

Restart your client if it does not discover the new tools automatically.

In Claude Code, use /mcp to check the connection. The command above installs OpenPulse for your user across projects.

Hermes configuration alternative: add this entry under mcp_servers in ~/.hermes/config.yaml, preserving existing servers.

mcp_servers:
  openpulse:
    url: https://openpulse.info/mcp
Tracked public repositories only. Unavailable data stays unavailable; queries do not trigger GitHub collection or AI analysis.Data & methodology ↗