Kizuna implements the Model Context Protocol (MCP) as a first-party server, exposing all forge operations as structured tools that any MCP-compatible agent can use.

What is MCP?

The Model Context Protocol (MCP) is an open standard for AI agent tool access. It enables:

• Structured tool definitions — Typed inputs/outputs
• Capability discovery — Agents see available tools
• Secure execution — Server validates all calls
• Broad compatibility — Works with Claude, Cursor, and more

Kizuna MCP Server

Kizuna's MCP server exposes the entire forge as tools:

{
  "tools": [
    {
      "name": "kizuna/create_change",
      "description": "Create a new change in a repository",
      "inputSchema": { ... }
    },
    {
      "name": "kizuna/open_pr",
      "description": "Open a pull request",
      "inputSchema": { ... }
    }
  ]
}

Connecting to MCP

Claude Desktop

Add to claude_desktop_config.json:

{
  "mcpServers": {
    "kizuna": {
      "command": "npx",
      "args": [
        "-y",
        "@kizuna/mcp-server",
        "--token",
        "YOUR_AGENT_TOKEN"
      ]
    }
  }
}

Claude Code

# Install Kizuna MCP
curl -sSL https://kizuna.codes/install-mcp | bash

# Configure with token
kizuna-mcp configure --token YOUR_AGENT_TOKEN

# Start
kizuna-mcp serve

Cursor

Add to Cursor settings:

{
  "mcpServers": {
    "kizuna": {
      "url": "https://kizuna.example.com/mcp",
      "headers": {
        "Authorization": "Bearer YOUR_AGENT_TOKEN"
      }
    }
  }
}

Available Tools

Repository Tools

Tool	Description
kizuna/clone	Clone a repository
kizuna/list_changes	List changes in repo
kizuna/get_change	Get specific change details
kizuna/create_change	Create new change
kizuna/describe_change	Update change description
kizuna/rebase_change	Rebase change to new parent
kizuna/squash_changes	Combine changes
kizuna/push_change	Push to remote

Pull Request Tools

Tool	Description
kizuna/open_pr	Create pull request
kizuna/get_pr	Get PR details
kizuna/list_prs	List open PRs
kizuna/comment_on_pr	Add review comment
kizuna/approve_pr	Approve pull request
kizuna/merge_pr	Merge pull request
kizuna/close_pr	Close without merging

Code Review Tools

Tool	Description
kizuna/get_diff	Get change diff
kizuna/get_file	Read file content
kizuna/search_code	Search repository
kizuna/get_blame	Get line annotations
kizuna/add_comment	Add inline comment

Agent Orchestration Tools

Tool	Description
kizuna/spawn_agent	Create sub-agent task
kizuna/get_agent_status	Check agent task status
kizuna/send_message	Send A2A message
kizuna/list_active_agents	See active agents

CI/CD Tools

Tool	Description
kizuna/trigger_pipeline	Run CI pipeline
kizuna/get_pipeline_status	Check pipeline status
kizuna/get_job_logs	Fetch CI logs
kizuna/approve_deployment	Approve deploy

Context Tools

Tool	Description
kizuna/get_repo_summary	Repository overview
kizuna/get_file_with_history	File + recent changes
kizuna/search_codebase	Semantic code search
kizuna/get_related_changes	Find similar changes
kizuna/read_intent	Read INTENT.md

Governance Tools

Tool	Description
kizuna/get_trust_level	Check agent's trust
kizuna/get_capabilities	List allowed actions
kizuna/get_audit_log	Query audit trail

Example Tool Calls

Create a Change

{
  "tool": "kizuna/create_change",
  "params": {
    "repo": "my-org/project",
    "parent": "main",
    "description": "Add error handling to auth module"
  }
}

Response:

{
  "change_id": "mnxpqkpv",
  "parent": "main",
  "description": "Add error handling to auth module",
  "url": "https://kizuna.example.com/repos/my-org/project/changes/mnxpqkpv"
}

Open a PR

{
  "tool": "kizuna/open_pr",
  "params": {
    "repo": "my-org/project",
    "title": "Add error handling",
    "body": "This PR adds try/catch blocks to auth endpoints",
    "head": "mnxpqkpv",
    "base": "main"
  }
}

Search Code

{
  "tool": "kizuna/search_code",
  "params": {
    "repo": "my-org/project",
    "query": "function authenticate",
    "language": "typescript"
  }
}

Tool Discovery

Agents can discover available tools:

{
  "tool": "kizuna/list_tools",
  "params": {}
}

Response includes all tools the agent's trust level allows.

Error Handling

Tools return structured errors:

{
  "error": {
    "code": "INSUFFICIENT_TRUST",
    "message": "Trust level 1 required for merge",
    "current_level": 0,
    "required_level": 1
  }
}

Common error codes:

• UNAUTHORIZED — Invalid credentials
• INSUFFICIENT_TRUST — Trust level too low
• RESOURCE_NOT_FOUND — Repo/change doesn't exist
• VALIDATION_ERROR — Invalid parameters
• RATE_LIMITED — Too many requests

Security

• Token-based auth: Each agent has unique credentials
• Trust enforcement: Tools check trust levels
• Audit logging: All calls logged
• Rate limiting: Prevents abuse
• Scope validation: Agents limited to capabilities

Next Steps

• A2A Communication — Inter-agent messaging
• Reference: MCP Tools — Complete tool reference