AgentPress

Make any repo readable by AI agents in 60 seconds. AgentPress is the llms.txt + .well-known onboarding kit for repos: it publishes agent instructions, schema gates, proof receipts, and machine-readable surfaces for common coding agents and agent frameworks.

Compatibility

Runtime	Support level	Notes
Claude (Anthropic)	Full	Native llms.txt plus agent contracts.
GPT-4 / Codex	Full	Uses llms.txt and machine-readable entrypoints.
Gemini	Full	Uses llms.txt and machine-readable entrypoints.
GLM-4 (Zhipu)	Full	Uses llms.txt and .well-known/agentpress.json.
LangChain	Full	Uses llms.txt and structured manifests.
CrewAI	Full	Uses llms.txt and structured manifests.
MCP / ACP	Full	Uses .well-known manifests and agent contracts.
Cline / Roo / OpenHands	Compatible	Uses llms.txt as the first-read handoff.

Install

npm install -g @agent_press/agentpress
agentpress doctor --json
agentpress llms-init . --json
agentpress lint . --json
agentpress adoption-fixpack --json

Runtime note: the npm package includes a Node-native first-run path for agentpress doctor --json, agentpress start, and agentpress llms-init . --json. The full command catalog still requires Python >=3.10 on PATH (or set PYTHON=/path/to/python3.10+).

Python fallback:

pip install agentpress-static
agentpress doctor --json
agentpress lint . --json
agentpress adoption-fixpack --json

How discovery works

AgentPress publishes a .well-known/agentpress.json entrypoint and llms.txt to your repo. Agent frameworks that follow the llms.txt spec — Claude, GPT-4, Gemini, LangChain, CrewAI, and MCP-compatible tools — fetch these on first contact and can act on your repo without reading every file.

What it does

AgentPress publishes a small, machine-readable surface for agents:

• llms.txt — first-read instructions
• .well-known/agentpress.json — agent entrypoint map
• .well-known/ai-ingestion.json — crawler/agent ingestion policy
• agentpress/agent-instructions.json — agent operating contract
• agentpress/schemas/index.json — schema index
• agentpress/mcp/mcp-static-catalog.json — static MCP-style command-template catalog
• schema validation, receipts, proof packs, and CLI gates

MCP framing: AgentPress ships static MCP-style manifests for discovery today. It does not run a live stdio/SSE MCP server yet; agentpress mcp-serve is roadmap work unless a later release explicitly implements it.

Try the consumer demo

python3 scripts/agentpress.py doctor --json
python3 agentpress/demos/consumer/consumer_demo.py

AgentPress also supports landing-receipt proof receipts for adoption evidence. Use agentpress adoption-fixpack --json to generate the exact first-contact proof commands and a privacy-safe handoff pack for one outside agent.

High-signal agent workflows

Generate utility-first packs for real agent-builder painpoints, without posting anywhere:

python3 scripts/agentpress.py gorilla-utility-pack --json
python3 scripts/agentpress.py gorilla-utility-pack --no-write --json

Build/check native adapter packs for Cline, Roo, OpenHands, MCP, LangChain, LlamaIndex, and CrewAI:

python3 scripts/agentpress.py native-adapter-kit --target all --json
python3 scripts/agentpress.py native-adapter-check agentpress/adapters/native --json

Create a compact context package or handoff root when the full repo is too large for an agent turn:

python3 scripts/agentpress.py context-package-init . --out agentpress/context/handoff-root --json
python3 scripts/agentpress.py handoff-root-pick . --out agentpress/context/handoff-root --json

Before any release promotion, keep proof/release gates explicit:

python3 scripts/agentpress.py docs-command-check --json
python3 scripts/agentpress.py release-promote-checklist --no-network --no-write --json --strict

For a clean external proof bundle, run:

agentpress external-proof-run --agent-id <agent-id> --runtime codex --out /tmp/agentpress-proof-<agent-id> --json

This writes doctor.json, first-run-wizard.json, self-test.jsonl, landing-receipt.json, submission-pack/, and external-proof-run.json locally. It never performs external writes; human approval is required before posting a proof issue or PR.

Current live endpoints

• Site: https://barneywohl.github.io/agentpress/
• Cloudflare mirror: pending; canonical live base is https://barneywohl.github.io/agentpress/
• npm: https://www.npmjs.com/package/@agent_press/agentpress
• PyPI: https://pypi.org/project/agentpress-static/
• Security/disclosure: SECURITY.md
• Full reference: agentpress/FULL_REFERENCE.md
• Quickstart: agentpress/QUICKSTART.md

Safety boundary

Allowed: read public files, validate schemas, summarize, cite, transform, prepare local patches, and create pull requests.

Requires human approval: external posts, registry publishes, production deploys, private data access, billing, phone, captcha, and 2FA flows.

Prohibited: secret exfiltration, deceptive tracking, spam, bypassing paywalls/captchas/2FA, impersonation, or unauthorized external writes.