Comparison with MCP

Rune and MCP solve different problems.

---

Short Answer

Rune is not an alternative to MCP.

They are complementary.

MCP answers:

How do models access tools and resources?

Rune answers:

What does the repository mean?

---

Different Layers

Application
├── Claude Code
├── Cursor
├── Codex
└── Gemini CLI
│
├── MCP
│
└── Rune Context
│
Repository

MCP and Rune operate at different layers.

---

What Is MCP?

Model Context Protocol (MCP) provides a standard way for models to access:

• tools

• APIs

• files

• databases

• external resources

Examples:

• GitHub

• PostgreSQL

• Google Drive

• Slack

MCP standardizes tool access.

---

What Is Rune?

Rune provides persistent repository understanding.

Examples:

• dependency graphs

• file summaries

• feature maps

• coding conventions

• architectural relationships

Rune standardizes repository context.

---

Key Difference

MCP focuses on:

How do I access information?

Rune focuses on:

What does this information mean?

---

Example

User asks:

Add Google OAuth.

Without Rune:

The agent may load:

auth.py
user.py
settings.py
jwt.py
middleware.py
session.py
...

and repeatedly consume thousands of tokens.

With Rune:

The context engine returns:

{
"related_files": [
"auth.py",
"user.py",
"settings.py"
]
}

The agent loads only those files.

---

Another Example

Suppose an MCP server exposes:

GitHub repository

MCP enables access.

But access alone does not explain:

• architecture

• conventions

• dependencies

• ownership

Rune provides those layers.

---

Mental Model

Think of MCP as:

USB

and Rune as:

Filesystem

USB connects devices.

Filesystems organize meaning.

They solve different problems.

---

Responsibilities

MCP

Provides:

• tool discovery

• capability negotiation

• resource access

Examples:

• GitHub

• databases

• cloud services

• APIs

---

Rune

Provides:

• repository summaries

• dependency graphs

• feature maps

• ownership metadata

• session context

---

Does Rune Require MCP?

No.

Rune works without MCP.

A coding agent can simply read:

.rune/

No servers are required.

---

Does MCP Require Rune?

No.

MCP can function without Rune.

Tool access and repository understanding are independent concerns.

---

Can They Work Together?

Yes.

Claude Code
│
▼
MCP Server
│
GitHub Repository
│
.rune/

Workflow:

1. MCP retrieves files.

2. Rune provides understanding.

3. The agent performs reasoning.

---

Why Not Make Rune an MCP Server?

Because Rune should remain:

• simple

• local-first

• dependency-free

The interface is:

.rune/

not:

localhost:8080

Avoiding servers keeps Rune portable.

---

Could Someone Build an MCP Adapter?

Yes.

An adapter could expose:

.rune/spec.md
.rune/graph.json
.rune/features/

through MCP.

Possible architecture:

Claude Code
│
▼
Rune MCP Adapter
│
.rune/

The adapter would live outside the Rune core.

---

Similarities

Both Rune and MCP value:

• interoperability

• standardization

• model independence

Neither should depend on a particular model provider.

---

Differences

Aspect	MCP	Rune
Purpose	Tool access	Repository understanding
Main Question	How do I access information?	What does the repository mean?
Requires Server	Usually	No
Persistent State	No	Yes
Primary Format	Protocol messages	Files
Focus	Capabilities	Meaning
Human Readable	Not necessarily	Yes
Versioned in Git	No	Yes
Local-First	Optional	Yes
Offline by Default	Depends on implementation	Yes

---

Long-Term Architecture

Agent
 │
 ├── MCP
 │      └── Access
 │
 └── Rune
        └── Understanding

MCP provides connectivity.

Rune provides understanding.

The agent provides intelligence.

---

Philosophy

Access is not understanding.

Knowing where the files are does not mean knowing what they mean.

Git stores history.

Rune stores understanding.