Model Context Protocol · 15+ tools

One config line.
Every agent, wired.

Add the CodeOrch MCP server to your .mcp.json, restart your agent, and 15+ orchestration tools are instantly available — persistent memory, intelligent briefings, file locking, review gates.

15+
MCP tools
~2s
briefing time
30s
to set up
3+
agents supported
§ 01  Configure30-second setup

From zero to first
orchestrated task.

01
Sign up

Create a free account. No card required.

02
Copy your API key

Generate a key from your project settings.

03
Add to .mcp.json

Two lines. Your agent picks it up on restart.

04
Call begin()

Your first briefing synthesizes in ~2 seconds.

MCP Configuration

Connect external agents to this project via Model Context Protocol.

claude_code_config.json
Claude Code

For the official Claude Code application (SSE transport). Replace YOUR_KEY_HERE with your server URL and a valid key from your account settings.

{
  "mcpServers": {
    "project-os": {
      "type": "sse",
      "url": "https://api.codeorch.app/mcp/sse",
      "headers": {
        "Authorization": "Bearer pos_xxx"
      }
    }
  }
}
§ 02  Agent rulesCLAUDE.md · .cursorrules · AGENTS.md

Tell your agent
how to behave.

Drop a rule file into your repo root. Claude reads CLAUDE.md. Cursor reads .cursorrules. Every agent gets the same baseline: claim before coding, read the briefing, finish with proof.

Agent Rules

Download rule files for Claude Code, Copilot, Cursor, and other AI agents.

CLAUDE.md
Claude Code

Default rules for Claude Code and Claude Desktop. Save this in your repository root to enable Claude Desktop integration.

# Claude Rules

- Always use the project-os MCP server for task management.
- Follow the instructions in get_task_briefing strictly.
- Maintain architectural integrity.
§ 03  MCP tools15+ available

Every tool your agent
needs to ship.

17 MCP tools organized across four stages — setup, task lifecycle, code navigation, and memory. Your agent calls them; the orchestrator handles the rest.

Project Setup
workspace_init

Auto-detects project, writes CLAUDE.md / .cursorrules / AGENTS.md, returns project ID and task state.

initialize_project

Name-only project creation — no cwd needed. Use from any directory.

list_projects

Returns all projects in your org with IDs, task counts, and last-activity timestamps.

Task Lifecycle
begin

Creates, claims, orchestrates, and starts a task in one call. Returns a briefing with files to focus.

finish

Submits work for review with structured proof. Auto-runs the 7-metric review gate.

create_task

Queue a task on the board without claiming. Useful for planning multi-agent pipelines.

get_task_briefing

Recovery tool — re-fetch a briefing if begin() was interrupted mid-session.

Code Navigation
find_files

Glob-safe file discovery against the indexed codebase. Never risks reading outside scope.

read_batch

Read up to 5 files in one call. Far faster than sequential file reads.

get_file_fresh

Re-reads a file after you've edited it — keeps your context consistent with disk.

search

Hybrid search across docs, decisions, and code. Hits graph, vector, and relational in parallel.

Planning & Memory
break_task

Decomposes a parent task into 3–5 child tasks. Orchestrator determines subtask scope.

plan_project

Takes a high-level goal and generates a full task board for multi-agent execution.

add_task_dependency

Links tasks so the orchestrator respects execution order across agents.

remember

Write decisions, session summaries, learnings, or docs directly to the knowledge graph.

reflect

Returns project health, stale items, decision gaps, and suggested next tasks.

sync

Clones a GitHub branch and re-indexes it. Accepts a branch argument.

Free tier forever

Give your agents
a memory.

Free tier. No card. Cloud or self-host. 100 orchestrations/month on the free plan — enough for 3–4 active tasks per day.

  • 2 projects included
  • 100 orchestrations / month
  • Claude Code, Cursor, Aider support
  • docker compose up for self-host