Metadata-Version: 2.3
Name: fastharness
Version: 0.2.1
Summary: Wrap Claude Agent SDK and expose agents as A2A-compliant FastAPI services
Author: Prass, The Nomadic Coder
Author-email: Prass, The Nomadic Coder <atemysemicolon@gmail.com>
Requires-Dist: claude-agent-sdk>=0.1.0
Requires-Dist: fasta2a>=0.1.0
Requires-Dist: fastapi>=0.115.0
Requires-Dist: pydantic>=2.0.0
Requires-Dist: uvicorn>=0.30.0
Requires-Dist: anyio>=4.0.0
Requires-Dist: pytest>=8.0.0 ; extra == 'dev'
Requires-Dist: pytest-asyncio>=0.24.0 ; extra == 'dev'
Requires-Dist: httpx>=0.27.0 ; extra == 'dev'
Requires-Dist: ty ; extra == 'dev'
Requires-Dist: ruff>=0.8.0 ; extra == 'dev'
Requires-Python: >=3.11
Provides-Extra: dev
Description-Content-Type: text/markdown

<p align="center">
  <img src="https://raw.githubusercontent.com/prassanna-ravishankar/fastharness/main/assets/logo.webp" alt="FastHarness" width="200">
</p>

# FastHarness

[![CI](https://github.com/prassanna-ravishankar/fastharness/actions/workflows/test.yml/badge.svg)](https://github.com/prassanna-ravishankar/fastharness/actions/workflows/test.yml)
[![PyPI](https://img.shields.io/pypi/v/fastharness)](https://pypi.org/project/fastharness/)
[![Python 3.11+](https://img.shields.io/badge/python-3.11+-blue.svg)](https://www.python.org/downloads/)
[![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](https://opensource.org/licenses/MIT)

Building agents with the Claude Agent SDK is straightforward, but exposing them as interoperable services requires implementing protocol layers, managing task lifecycles, and handling message conversion between formats. FastHarness bridges this gap by wrapping the Claude Agent SDK and automatically exposing your agents through Google's [A2A (Agent-to-Agent)](https://a2a-protocol.org) protocol.

Define agents with a simple decorator-based API. FastHarness handles the rest: generating agent cards, exposing JSON-RPC endpoints, converting between Claude SDK messages and A2A format, managing async task execution, and optionally tracking costs and logging intermediate steps. A simple agent requires only a name, description, and list of skills. For complex workflows, the `@agentloop` decorator gives you full control over the execution loop while FastHarness manages the protocol machinery.

Features include:
- **Cost tracking** - Monitor API usage with configurable thresholds via `CostTracker`
- **Step logging** - Capture tool calls, messages, and turn metrics with `ConsoleStepLogger`
- **CLAUDE.md support** - Automatically load project context and conventions
- **MCP servers** - Connect external services via Model Context Protocol
- **FastAPI integration** - Run standalone or mount on existing applications
- **LiteLLM support** - Route API calls through alternative providers

FastHarness is production-ready and fully A2A protocol compliant.

## Installation

```bash
uv add fastharness
```

## Quick Start

```python
from fastharness import FastHarness, Skill

harness = FastHarness(name="my-agent")

harness.agent(
    name="assistant",
    description="A helpful assistant",
    skills=[Skill(id="help", name="Help", description="Answer questions")],
    system_prompt="You are helpful.",
    tools=["Read", "Grep"],
)

app = harness.app
```

Run with:
```bash
uvicorn mymodule:app --port 8000
```

Test:
```bash
curl http://localhost:8000/.well-known/agent-card.json
```

## Custom Agent Loop

```python
@harness.agentloop(
    name="researcher",
    description="Multi-turn researcher",
    skills=[Skill(id="research", name="Research", description="Deep research")],
)
async def researcher(prompt, ctx, client):
    result = await client.run(prompt)
    while "need more" in result.lower():
        result = await client.run("Continue researching")
    return result
```

## Mount on FastAPI

```python
from fastapi import FastAPI
from contextlib import asynccontextmanager

@asynccontextmanager
async def lifespan(app):
    async with harness.lifespan_context():
        yield

app = FastAPI(lifespan=lifespan)
app.mount("/agents", harness.app)
```

## v1.0.0 Features

### CLAUDE.md Support

Agents automatically load project settings from `CLAUDE.md` by default:

```python
harness.agent(
    name="reviewer",
    description="Code reviewer",
    skills=[...],
    system_prompt="You are a thorough code reviewer.",
    # setting_sources=["project"] is default - loads CLAUDE.md automatically
)

# To disable, pass empty list:
harness.agent(
    name="readonly",
    description="Read-only assistant",
    skills=[...],
    setting_sources=[],  # Don't load any settings
)
```

### MCP Server Integration

Connect external services via Model Context Protocol:

```python
harness.agent(
    name="assistant",
    description="Multi-tool assistant",
    skills=[...],
    mcp_servers={
        "filesystem": {
            "command": "node",
            "args": ["mcp-server-stdio-filesystem"],
        },
    },
    tools=["mcp__filesystem__read", "mcp__filesystem__write"],
)
```

### Cost Tracking

Monitor API costs with configurable thresholds:

```python
from fastharness import CostTracker

tracker = CostTracker(warn_threshold_usd=1.0)

@harness.agentloop(...)
async def agent(prompt, ctx, client):
    client.telemetry_callbacks.append(tracker)
    result = await client.run(prompt)
    print(f"Cost: ${tracker.total_cost_usd:.4f}")
    return result
```

### Step Logging

Log intermediate steps for debugging:

```python
from fastharness import ConsoleStepLogger

client = HarnessClient(
    step_logger=ConsoleStepLogger(),
)
result = await client.run(prompt)
```

Output example:
```
[step_logger] Tool call
    turn: 0
    tool_name: Read
    tool_id: call_123
[step_logger] Assistant message
    turn: 0
    text_preview: Found the bug in line 42...
[step_logger] Turn complete
    turn: 0
    cost_usd: 0.01
    tokens: {'input': 150, 'output': 50}
```

## HarnessClient Options

The `HarnessClient` passed to agent functions supports these options:

| Option | Default | Description |
|--------|---------|-------------|
| `system_prompt` | `None` | System prompt for Claude |
| `tools` | `[]` | Allowed tools (e.g., `["Read", "Grep", "Glob"]`) |
| `model` | `claude-sonnet-4-20250514` | Claude model to use |
| `max_turns` | `None` | Maximum conversation turns |
| `permission_mode` | `bypassPermissions` | Permission handling mode |
| `mcp_servers` | `{}` | MCP server configurations |
| `setting_sources` | `["project"]` | Filesystem settings to load (loads CLAUDE.md) |
| `telemetry_callbacks` | `[]` | Callbacks for execution metrics |
| `step_logger` | `None` | Logger for intermediate steps |
| `output_format` | `None` | JSON schema for structured output (e.g., `{"type": "json_schema", "schema": {...}}`) |

Override per-call:
```python
result = await client.run(prompt, model="claude-opus-4-20250514", max_turns=5)
```

## A2A Endpoints

Running FastHarness exposes these endpoints:

| Endpoint | Description |
|----------|-------------|
| `/.well-known/agent-card.json` | Agent metadata and capabilities |
| `/` | JSON-RPC endpoint (`message/send`, `tasks/get`, etc.) |
| `/docs` | Interactive documentation |

## Examples

See the `examples/` directory for complete working examples:

- **[simple_agent.py](examples/simple_agent.py)** - Standalone agent service with multi-agent support
- **[fastapi_integration.py](examples/fastapi_integration.py)** - Mounting FastHarness on existing FastAPI app
- **[advanced_features.py](examples/advanced_features.py)** - Cost tracking, step logging, and MCP server configuration

Run examples:
```bash
# Standalone agent
uv run uvicorn examples.simple_agent:app --port 8000

# FastAPI integration
uv run uvicorn examples.fastapi_integration:app --port 8000
```

Test with:
```bash
# Get agent card
curl http://localhost:8000/.well-known/agent-card.json

# Send message (A2A JSON-RPC)
curl -X POST http://localhost:8000/ \
  -H "Content-Type: application/json" \
  -d '{
    "jsonrpc": "2.0",
    "method": "message/send",
    "params": {
      "id": "task-001",
      "contextId": "ctx-001",
      "message": {
        "role": "user",
        "parts": [{"kind": "text", "text": "Hello"}],
        "kind": "message",
        "messageId": "msg-001"
      }
    },
    "id": 1
  }'
```

## LiteLLM Support

Set environment variables to use LiteLLM as a proxy:

```bash
ANTHROPIC_BASE_URL=http://localhost:4000
ANTHROPIC_API_KEY=your-litellm-key
ANTHROPIC_MODEL=sonnet-4
```

## License

MIT
