# Osiris

> Osiris is an open-source MCP server for ai models and llm tooling. Context governor for coding agents: sealed task state, loop guard and payload offloading. It runs locally on your machine and is started by your AI client.

- **Kind:** MCP server
- **Pricing:** Open source
- **Platforms:** CLI, self-hosted, macOS, Windows, Linux
- **Repository:** https://github.com/mlkblk-it/osiris
- **GitHub stars:** 2
- **License:** MIT
- **MCP Registry name:** io.github.MlkBlk-IT/osiris
- **Categories:** [AI models and LLM tooling](https://ai-powered.com/categories/ai-and-llm), [developer tools](https://ai-powered.com/categories/developer-tools)
- **Listing updated:** 2026-11-18

## About

OSIRIS starts from an honest constraint: an MCP server never sees the final prompt, only what the agent sends it. So instead of editing the prompt, it produces a short deterministic manifest of about 120 tokens that the agent carries, sealed with a SHA-256 digest. On each turn the agent passes the seal back, and a mismatch reveals state drift. Separate hashes for head, plan and tail zones show whether a change breaks the prompt cache or merely appends.

Six tools cover the loop: plan, precheck, observe, recall, checkpoint and handoff. Large payloads are stored by reference and recalled with server-side grep, range, head, tail, outline, aggregate or jsonl filters so only matches reach the context. Other modules detect repeated-action loops (warn at 3, block at 5), enforce invariants from a TOML file, scrub secrets before hashing and keep a ledger of tokens, cost and savings. The tool schemas cost about 1,019 tokens.

It is local-first and independent of any model provider, stores state in SQLite under ~/.osiris, and is published on npm (osiris-mcp) and PyPI (mcp-osiris). It works alongside other MCP servers rather than replacing them.

## Features

- **Sealed manifest:** A small SHA-256-sealed state block that exposes drift.
- **Cache-aware hashing:** Tells cache-breaking changes from harmless appends.
- **Server-side recall:** Filters stored payloads with grep, range, outline and more.
- **Loop guard:** Detects repeated actions and escalates from warning to block.
- **Invariants and budgets:** Allows or denies actions based on rules and budgets.
- **Honest ledger:** Records tokens, cost, savings and its own overhead.
- **Delegation contracts:** Hands off tasks with executable contracts and structured results.

## Good for

- Keep a long coding task consistent across context resets
- Cut token use from large tool outputs
- Stop an agent repeating the same failing action
- Hand a subtask to another agent with a clear contract

## Install

### Claude Code (Terminal)

```sh
claude mcp add osiris -- npx -y osiris-mcp
```

### Claude Desktop (claude_desktop_config.json)

```json
{
  "mcpServers": {
    "osiris": {
      "command": "npx",
      "args": [
        "-y",
        "osiris-mcp"
      ]
    }
  }
}
```

### Cursor (~/.cursor/mcp.json)

```json
{
  "mcpServers": {
    "osiris": {
      "command": "npx",
      "args": [
        "-y",
        "osiris-mcp"
      ]
    }
  }
}
```

### VS Code (.vscode/mcp.json)

```json
{
  "servers": {
    "osiris": {
      "type": "stdio",
      "command": "npx",
      "args": [
        "-y",
        "osiris-mcp"
      ]
    }
  }
}
```

### Windsurf (~/.codeium/windsurf/mcp_config.json)

```json
{
  "mcpServers": {
    "osiris": {
      "command": "npx",
      "args": [
        "-y",
        "osiris-mcp"
      ]
    }
  }
}
```

### Other clients (your client's MCP settings (most use this shape))

```json
{
  "mcpServers": {
    "osiris": {
      "command": "npx",
      "args": [
        "-y",
        "osiris-mcp"
      ]
    }
  }
}
```

## Tools

- `osiris.plan`: Starts, updates, resumes or reports on a task and returns the sealed manifest.
- `osiris.precheck`: Authorizes or denies an action against invariants and budget.
- `osiris.observe`: Logs actions, detects loops and offloads large data.
- `osiris.recall`: Rehydrates references with server-side filtering.
- `osiris.checkpoint`: Snapshots, resumes and compacts a session.
- `osiris.handoff`: Creates delegation contracts and structured results.

## FAQ

### Does it replace other MCP servers?

No. It coexists with servers such as filesystem, Git and shell and reduces what they cost in context.

### Can it edit my prompt?

No. The README says an MCP server cannot see or rewrite the final prompt; it provides state the agent carries.

### How many tools does it expose?

Six: plan, precheck, observe, recall, checkpoint and handoff.

### Where does it keep data?

In ~/.osiris by default, set with OSIRIS_HOME, using SQLite.

### Is Osiris free?

Yes. Osiris is open source, so you can use and self-host it for free.

### How do I add Osiris to Claude Code?

Run this command in your terminal: claude mcp add osiris -- npx -y osiris-mcp

### What are the best alternatives to Osiris?

The closest alternatives to Osiris are REA, Context7, Chrome DevTools MCP, Codebase Memory and GitHub.

## Where this comes from

- Description: written from the project's README.
- Install steps: from the official MCP Registry entry.

## Alternatives

- [REA](https://ai-powered.com/mcp/rea): REA is an open-source MCP server for developer tools.
- [Context7](https://ai-powered.com/mcp/context7): Context7 is a freemium MCP server for developer tools.
- [Chrome DevTools MCP](https://ai-powered.com/mcp/chrome-devtools-mcp-2): Chrome DevTools MCP is an open-source MCP server for browser automation.
- [Codebase Memory](https://ai-powered.com/mcp/codebase-memory): Codebase Memory is an open-source MCP server for knowledge and memory.
- [GitHub](https://ai-powered.com/mcp/github-2): GitHub is an open-source MCP server for developer tools.
- [Serena MCP: the IDE for your agent](https://ai-powered.com/mcp/serena-mcp-the-ide-for-your-agent): Serena MCP: the IDE for your agent is an open-source MCP server for developer tools.
- [awesome-mcp-servers](https://ai-powered.com/mcp/awesome-mcp-servers-2): awesome-mcp-servers is an open-source MCP server for developer tools.
- [XERJ](https://ai-powered.com/mcp/xerj): XERJ is an open-source MCP server for knowledge and memory.

---
Source: https://ai-powered.com/mcp/osiris
