Built with ♥ for AI coding agents

The Documentation Layer
for AI Coding Agents

Give your AI coding agent accurate, up‑to‑date library documentation - so the code it writes actually works. No more hallucinated APIs. No more stale code.

Built on MCP. Works with Claude Code, Cursor, Windsurf, Cline, GitHub Copilot, OpenAI Codex, Antigravity, and any MCP‑compatible agent.

Get Started Free View on GitHub
~100% accuracy - limited only by source docs
2,000+ libraries indexed
84% of devs use AI tools†
free community edition
AI assistant + ProContext
# ── without ProContext ──────────────────
write a FastAPI route using Pydantic v2
AI using training data... knowledge cutoff: Oct 2023
class Item(BaseModel): name = str # ← v1 syntax price = float item.schema() # ← removed in v2
AttributeError: 'ModelMetaclass' has no
attribute 'schema'
# ── with ProContext ─────────────────────
write a FastAPI route using Pydantic v2
29%
of developers trust
AI output accuracy†

84% of developers use AI coding tools - yet trust in their accuracy has dropped to an all-time low. The #1 frustration: AI solutions that are almost right, but not quite. LLM frameworks ship breaking changes faster than any model can learn. You're asking an AI to write code for a world it hasn't seen yet.

† Stack Overflow Developer Survey 2025 · 49,000+ respondents

⏰
Stale Knowledge
AI training has a hard cutoff. LangChain 0.1 → 0.3, OpenAI SDK v0 → v1, LlamaIndex rewrites - your AI learned the old version and has no way to know it changed.
👻
Hallucinated APIs
AI invents method names, parameters, and class signatures that don't exist in any version. The code compiles. The runtime crashes.
🔥
Wasted Debugging Time
You spend 30+ minutes tracing a bug only to discover the AI used a method deprecated two major versions ago. Time you'll never get back.
How It Works

An Open-Book Exam
for Your AI Assistant

Without ProContext, your AI works from memory - like a closed-book exam. Memory goes stale. Answers go wrong. ProContext gives it the reference material it needs, exactly when it needs it.

STEP 01
📦

Resolve

Finds the right library and version from your query or imports. 2,000+ libraries indexed and ready.

→
STEP 02
🔍

Search

Finds the exact section that's relevant - no need to know the exact name or where it lives.

→
STEP 03
📖

Read

Reads the full documentation or any specific page your agent needs - always the current version.

→
STEP 04
🗂️

Outline

Maps the structure of available docs so your agent knows where to look before diving in.

// zero_friction = true ProContext works entirely behind the scenes. No workflow changes. No prompts to write. Install it once and forget it's there.
Setup

Install Once.
Works Everywhere.

Free and open source. No daemons. No config files to learn. No API keys. Run one command to install, add one config block, and you're done.

1
Install
macOS / Linux
$ curl -fsSL https://procontext.dev/install.sh | bash
Windows (PowerShell)
> irm https://procontext.dev/install.ps1 | iex
2
Add to your MCP client config
~/.cursor/mcp.json  ·  or any MCP client
{
  "mcpServers": {
    "procontext": {
      "command": "uvx",
      "args": ["procontext"]
    }
  }
}
3
Start coding
Your AI assistant now queries ProContext automatically on every code generation. No prompts to write. No workflow changes. You won't notice it's there - until you stop getting runtime errors.
Community Edition

Free & Open Source

Everything you need to give your AI coding agent accurate, up-to-date library documentation. Self-host on your own infra, MIT licensed.

  • ✓ 2,000+ community libraries indexed
  • ✓ 4 core MCP tools — resolve, search, read, outline
  • ✓ Public custom sources
  • ✓ stdio transport + local cache
  • ✓ Self-host on your own infra
  • ✓ MIT open source
  • ✓ Community support (GitHub)
Download on GitHub
Compatibility

Works With Every
AI Coding Assistant

ProContext is built on the MCP standard. If your AI client supports MCP, ProContext works with it - no custom integrations needed.

AI Coding Assistants
Claude Code
OpenAI Codex
Cursor
GitHub Copilot
Antigravity
Any MCP Client
Operating Systems
macOS
Linux
Windows
Built on latest MCP spec  ·  stdio & Streamable HTTP transport  ·  Python 3.12+
Why ProContext

Open-Source.
Without the Accuracy Trade-off.

High-accuracy documentation tools are proprietary. Open-source options sacrifice accuracy. ProContext fills that gap - open-source, high accuracy, free for individuals, SaaS for teams.

✗
Context7
~60% accuracy in third-party benchmarks*  ·  Proprietary  ·  Cloud-only  ·  No self-hosting
✗
Deepcon
~90% accuracy in third-party benchmarks*  ·  Closed source  ·  No community edition  ·  Vendor lock-in
✗
Other open-source tools
<50% accuracy  ·  No pre-processing  ·  No dual retrieval modes
✓
ProContext
Accuracy tied to official source docs  ·  MIT open source  ·  SaaS & self-hosted  ·  Free community edition
↑ ACCURACY
HIGH ACCURACY
PROPRIETARY
Deepcon
90%
HIGH ACCURACY
OPEN SOURCE
ProContext
source-accurate
LOW ACCURACY
PROPRIETARY
Context7
~60%*
LOW ACCURACY
OPEN SOURCE
Others
<50%
← Proprietary Open Source →
OPENNESS

* Accuracy figures sourced from third-party benchmarks. Results may vary by benchmark methodology and use case. See published comparative analysis.  ·  † Stack Overflow Developer Survey, 2025.