Ivy Tendril Developer Portal

1. Model Context Protocol (MCP)

Connect Claude Desktop, ChatGPT, or custom agent frameworks to the built-in MCP server:

# Stdio transport:
tendril mcp

# Streamable HTTP / SSE transport (via background daemon):
http://127.0.0.1:5010/mcp

Discovery endpoint: /.well-known/mcp.json | Documentation: MCP Guide

2. Tendril Command Line Interface (CLI)

Supervise and execute plans directly from your shell:

# Start the background daemon
tendril run

# Submit and run a plan in an isolated git worktree
tendril plan create "Add OAuth2 PKCE flow"
tendril plan run <plan-id>

# Check environment health
tendril doctor

Zero Friction, Local-First Key Management

Ivy Tendril is 100% Free and Source-Available under the Functional Source License (FSL-1.1-ALv2). There are no credit cards, paywalls, or "contact sales" forms.

Generate and store API keys directly on your device using Tendril's local encrypted keyring:

# Initialize your local vault
tendril vault init

# Set model provider API keys locally (never transmitted to Ivy servers)
tendril config set-key anthropic <your-key>
tendril config set-key openai <your-key>
tendril config set-key openrouter <your-key>

For HTTP daemon authentication, the daemon issues local Bearer tokens scoped to plans:read, plans:write, worktrees:read, and worktrees:write. Inspect discovery metadata at /.well-known/oauth-authorization-server.

Safe Agent Testing with Mock Daemon

Test your agent workflows, MCP integrations, and plan execution loops without touching real repositories or burning LLM tokens:

# Start Tendril daemon in sandbox mode:
tendril run --sandbox

# Access sandbox endpoints on port 5010:
curl -s http://127.0.0.1:5010/sandbox/plans

In sandbox mode:

• All git worktrees are ephemeral and created in temporary memory directories.
• Verification test runners execute deterministic mock suites (unit, lint, visual).
• Agent responses can be replayed from pre-recorded deterministic fixtures.