CPE – Chat-based Programming Editor

---

CPE connects your local development workflow to multiple AI providers through a single, unified interface. Write natural language prompts, and CPE handles the rest—whether you're analyzing code, making edits, or automating complex tasks.

✨ Why CPE?

• One tool, many models: Switch between Claude, GPT, Gemini, and more with a simple flag
• Tool integration: Connect to MCP servers for file editing, shell commands, web search, and more
• Conversation memory: Resume previous conversations or branch off in new directions
• Code Mode: Let the AI write and execute Go code to accomplish complex multi-step tasks
• Privacy-first: Your data stays local; incognito mode for sensitive work

🚀 Installation

Using release binaries

curl -fsSL https://raw.githubusercontent.com/spachava753/cpe/main/install.sh | sh

The installer downloads the latest release for macOS or Linux and installs cpe to ~/.local/bin by default. Set CPE_INSTALL_DIR=/usr/local/bin or CPE_INSTALL_VERSION=vX.Y.Z to customize the install.

Using Go

go install github.com/spachava753/cpe@latest

From source

git clone https://github.com/spachava753/cpe.git
cd cpe
go build -o cpe .

Shell Completion

CPE supports shell autocompletion for faster command entry:

# Bash (add to ~/.bashrc)
source <(cpe completion bash)

# Zsh (add to ~/.zshrc)
source <(cpe completion zsh)

# Fish
cpe completion fish | source

# PowerShell
cpe completion powershell | Out-String | Invoke-Expression

⚡ Quick Start

Note: CPE requires a configuration file to define which models and tools to use. There's no zero-config mode—you'll need to set up at least one model before getting started.

1. Create a configuration file

Create a cpe.yaml in your project directory or in your user config directory:

• macOS: ~/Library/Application Support/cpe/cpe.yaml
• Linux: ~/.config/cpe/cpe.yaml
• Windows: %AppData%\cpe\cpe.yaml

version: "1.0"

models:
  - ref: sonnet
    display_name: "Claude Sonnet"
    id: claude-sonnet-4-5-20250929
    type: anthropic
    api_key_env: ANTHROPIC_API_KEY  # You choose the env var name
    context_window: 200000
    max_output: 64000
    timeout: 5m
    generationParams:
      temperature: 0.2

Tip: You can quickly add models from the models.dev registry:

# Add a model from the registry
cpe config add anthropic/claude-sonnet-4-20250514 --ref sonnet

2. Set your API key

# Use whatever env var name you specified in api_key_env
export ANTHROPIC_API_KEY="your-api-key"

3. Start chatting

# Ask a question (or set CPE_MODEL=sonnet once in your shell)
cpe --model sonnet "Explain what this project does"

# Analyze specific files
cpe -i main.go -i README.md "What are the main entry points?"

# Use a different model
cpe -m gpt4 "Help me refactor this function"

🎯 Features

Multi-Model Support

CPE works with all major AI providers:

Provider	Type	Example Models
Anthropic	anthropic	Claude Opus, Sonnet, Haiku
OpenAI	openai, responses	GPT-4o, GPT-5, o1
Google	gemini	Gemini Pro, Flash
Groq	groq	Llama, Mixtral (fast inference)
Z.AI (ZhipuAI)	zai	GLM-4
Cerebras	cerebras	Llama (fast inference)
OpenRouter	openrouter or openai (with base_url)	Any OpenRouter model

Switch models on the fly:

cpe -m sonnet "Write a test for this function"
cpe -m flash "Quick question about Go syntax"

Conversation Persistence

CPE remembers your conversations, stored locally in .cpeconvo:

# Continue your last conversation
cpe "Now add error handling to that code"

# Start fresh
cpe -n "Let's work on a new feature"

# Continue from a specific point
cpe -c abc123 "Actually, let's try a different approach"

# View conversation history (aliases: convo, conv)
cpe conversation list       # alias: ls

# Print a specific conversation (aliases: show, view)
cpe conversation print abc123

# Delete a conversation (aliases: rm, remove)
cpe conversation delete abc123

# Delete with cascade (removes children too)
cpe conversation delete abc123 --cascade

Input Files & URLs

Feed context directly to CPE:

# Include files
cpe -i src/main.go -i src/utils.go "Find any bugs"

# Include URLs
cpe -i https://example.com/docs.md "Summarize this documentation"

# Mix files and stdin
echo "Additional context" | cpe -i config.yaml "Help me configure this"

MCP Tool Integration

Connect external tools via the Model Context Protocol or enable bundled tools with type: builtin. CPE supports four server types:

Type	Description	Use Case
builtin	CPE-provided in-process tools	File editing without installing an MCP server
stdio	Local process via stdin/stdout	Local tools, CLIs
http	HTTP/HTTPS endpoint	Remote APIs, cloud services
sse	Server-Sent Events	Streaming, real-time tools

# cpe.yaml
models:
  - ref: sonnet
    display_name: "Claude Sonnet"
    id: claude-sonnet-4-5-20250929
    type: anthropic
    api_key_env: ANTHROPIC_API_KEY
    context_window: 200000
    max_output: 64000
    mcpServers:
      # Built-in file editing tool, no external MCP server required
      editor:
        type: builtin
        timeout: 60
        enabledTools:
          - text_edit

      # Remote tool via HTTP
      search:
        url: "https://search.example.com/mcp"
        type: http
        headers:
          Authorization: "Bearer ${API_KEY}"

      # SSE-based server
      streaming:
        url: "https://streaming.example.com/sse"
        type: sse
        headers:
          X-API-Key: "${MCP_API_KEY}"

Tool Filtering: Control which tools are exposed to the AI:

• enabledTools: Whitelist—only these tools are available
• disabledTools: Blacklist—all tools except these are available

Now the AI can edit files, run commands, and search the web!

Debugging MCP Integrations

CPE provides commands to help debug MCP server connections:

# List MCP servers for the selected model profile (alias: ls-servers)
cpe --model sonnet mcp list-servers

# List tools available from a specific server (alias: ls-tools)
cpe --model sonnet mcp list-tools editor

# Show all tools including filtered ones
cpe --model sonnet mcp list-tools editor --show-all

# Show only filtered-out tools
cpe --model sonnet mcp list-tools editor --show-filtered

# Get detailed info about a server
cpe --model sonnet mcp info editor

# Call a tool directly for testing
cpe --model sonnet mcp call-tool --server editor --tool text_edit --args '{"path": "test.txt", "text": "hello"}'

# View the execute_go_code tool description (for code mode)
cpe --model sonnet mcp code-desc

Code Mode

Code Mode lets the AI write and execute Go code to accomplish complex tasks in a single step:

models:
  - ref: sonnet
    # ...model provider fields...
    codeMode:
      enabled: true
      localModulePaths:
        - ../my-go-helpers
        - /Users/me/dev/shared-go-utils

With Code Mode, the AI can:

• Chain multiple tool calls together
• Use loops and conditionals
• Process data in parallel
• Access Go's standard library

When localModulePaths is configured, generated code runs inside an ephemeral Go workspace. This lets go mod tidy, go build, and import auto-correction resolve local modules without per-run manual setup.

Builtin MCP tools such as text_edit are intentionally not exposed inside Code Mode. They remain normal conversational tools even when Code Mode is enabled.

Example: "Find all TODO comments, group them by file, and create a summary report" becomes a single Go program that the AI writes and CPE executes.

Account Authentication

Use your Claude Pro/Max or ChatGPT subscription directly:

# Authenticate with Anthropic or OpenAI
cpe account login anthropic
cpe account login openai

# Show OpenAI subscription usage
cpe account usage openai

# Watch usage live
cpe account usage openai --watch

# Get the raw JSON response for scripts
cpe account usage openai --raw

# Override the ChatGPT usage backend base URL
cpe account usage openai --base-url https://chatgpt.com/backend-api

# Logout
cpe account logout anthropic
cpe account logout openai

Request Patching

For advanced use cases, you can patch API requests with custom headers or JSON modifications. This is useful for:

• OpenRouter's required HTTP-Referer header
• Custom authentication schemes
• Adding provider-specific metadata

models:
  - ref: qwen
    id: qwen/qwen3-max
    type: openai
    base_url: https://openrouter.ai/api/v1/
    api_key_env: OPENROUTER_API_KEY
    patchRequest:
      includeHeaders:
        HTTP-Referer: https://my-app.example.com
        X-Title: My AI App
      # Optional: JSON Patch operations
      # jsonPatch:
      #   - op: add
      #     path: /custom_field
      #     value: custom_value

⚙️ Configuration

Configuration File Locations

CPE searches for configuration in this order:

1. --config flag (explicit path)
2. ./cpe.yaml (current directory)
3. User config directory:
   • macOS: ~/Library/Application Support/cpe/cpe.yaml
   • Linux: ~/.config/cpe/cpe.yaml
   • Windows: %AppData%\cpe\cpe.yaml

Full Configuration Example

# yaml-language-server: $schema=https://raw.githubusercontent.com/spachava753/cpe/refs/heads/main/schema/cpe-config-schema.json
version: "1.0"

# Define complete model profiles. Use YAML anchors to avoid duplication.
models:
  - ref: sonnet
    display_name: "Claude Sonnet"
    id: claude-sonnet-4-5-20250929
    type: anthropic
    api_key_env: ANTHROPIC_API_KEY
    context_window: 200000
    max_output: 64000
    input_cost_per_million: 3
    output_cost_per_million: 15
    mcpServers: &standardMCPServers
      editor:
        type: builtin
        timeout: 60
        enabledTools:
          - text_edit

  - ref: flash
    display_name: "Gemini Flash"
    id: gemini-flash-latest
    type: gemini
    api_key_env: GEMINI_API_KEY
    context_window: 1048576
    max_output: 65536

  - ref: gpt4
    display_name: "GPT-4o"
    id: gpt-4o
    type: openai
    api_key_env: OPENAI_API_KEY
    context_window: 128000
    max_output: 16384

  - ref: glm
    display_name: "Z.AI GLM-4"
    id: glm-4
    type: zai
    api_key_env: Z_API_KEY
    context_window: 128000
    max_output: 4096

    systemPromptPath: "./prompts/agent.md"
    timeout: 5m
    codeMode:
      enabled: true
      maxTimeout: 3600
      localModulePaths:
        - ../my-go-helpers
    # Generation parameters control LLM behavior for this profile.
    generationParams:
      temperature: 0.7      # Controls randomness (0.0 = deterministic, 1.0 = creative)
      # topP: 0.9           # Nucleus sampling threshold
      # topK: 40            # Top-k sampling parameter
      # frequencyPenalty: 0 # Penalize repeated tokens (-2.0 to 2.0)
      # presencePenalty: 0  # Penalize tokens already in context (-2.0 to 2.0)

Environment Variables

Variable	Description
CPE_MODEL	Model profile to use when --model is not passed
CPE_DB_PATH	Conversation SQLite database path when --db-path is not passed

Note: API keys are configured per-model via the api_key_env field. You choose the environment variable name—there are no hardcoded defaults. For example, you could use MY_ANTHROPIC_KEY, OPENAI_API_KEY, or any name you prefer.

Model Management

# List configured models (aliases: models, ls)
cpe model list

# Show model details
cpe model info sonnet

# View the rendered system prompt for a model (uses -m flag)
cpe model system-prompt -m sonnet

# Add a model from models.dev registry
cpe config add anthropic/claude-sonnet-4-20250514

# Add with custom ref
cpe config add anthropic/claude-sonnet-4-20250514 --ref claude

# Remove a model
cpe config remove claude

📚 Examples

Code Review

cpe -i main.go -i utils.go "Review this code for potential bugs and suggest improvements"

Refactoring

cpe -i legacy_module.py "Refactor this to use modern Python patterns. Update the file directly."

Documentation

cpe -i api.go -i handlers.go "Generate comprehensive documentation for all public functions"

Debugging

cpe -i error.log -i src/handler.go "Why is this error happening? Suggest a fix."

Quick Tasks

# Generate a .gitignore
cpe "Create a .gitignore for a Go project"

# Explain code
cpe -i complex_algorithm.go "Explain what this does step by step"

# Convert formats
echo '{"name": "test"}' | cpe "Convert this JSON to YAML"

🧩 Skills System

CPE can be extended with skills—reusable, composable capabilities that provide specialized knowledge and workflows.

What Are Skills?

A skill is a directory containing a SKILL.md file with YAML frontmatter (name and description) followed by instructions, examples, or reference material. Skills are discovered and rendered into the system prompt.

Configuring Skills

Skills locations are user-defined in your system prompt template using the {{ skills }} function. There are no hardcoded defaults—you specify exactly which directories to scan:

<!-- In your agent_instructions.md template -->
{{- $skills := skills "./skills" "~/my-custom-skills" "/shared/team-skills" -}}
{{- if $skills }}
<skills>
{{- range $skill := $skills }}
  <skill name={{ printf "%q" $skill.Name }}>
    <description>{{ $skill.Description }}</description>
    <path>{{ $skill.Path }}</path>
  </skill>
{{- end }}
</skills>
{{- end }}

The skills function:

• Accepts any number of directory paths
• Scans each for subdirectories containing SKILL.md
• Returns a list of skill objects (name, description, path) so your template controls the output format (XML, JSON, CSV, etc.)

Example Skill Structure

skills/
└── github-issue/
    └── SKILL.md

---
name: github-issue
description: Create and manage GitHub issues with proper templates
---

# GitHub Issue Skill

Instructions for creating well-formatted GitHub issues...

Creating Your Own Skills

Create a directory with a SKILL.md file anywhere you like, then reference that path in your system prompt template:

{{- $skills := skills "./my-project-skills" "~/my-global-skills" -}}
{{/* format $skills however you want */}}

For examples of well-structured skills, see the skills/ directory in the CPE repository—these are skills used for CPE's own development but serve as good templates for creating your own.

🔧 CLI Reference

cpe [flags] [prompt]

Core Flags:
  -m, --model string           Specify the model profile to use (required unless CPE_MODEL is set)
  -i, --input strings          Input files or URLs to process
  -n, --new                    Start a new conversation
  -c, --continue string        Continue from a specific conversation ID
  -G, --incognito              Don't save conversation to storage
      --config string          Path to YAML configuration file
      --db-path string         Path to conversation SQLite database
      --skip-stdin             Skip reading from stdin
  -v, --version                Print version and exit

Generation Parameters:
  -t, --temperature float      Sampling temperature (0.0 - 1.0)
  -x, --max-tokens int         Maximum tokens to generate
  -b, --thinking-budget string Budget for reasoning capabilities
      --top-p float            Nucleus sampling parameter (0.0 - 1.0)
      --top-k uint             Top-k sampling parameter
      --frequency-penalty float Frequency penalty (-2.0 - 2.0)
      --presence-penalty float  Presence penalty (-2.0 - 2.0)
      --number-of-responses uint Number of responses to generate
      --timeout string         Request timeout (e.g., '5m', '30s')

Advanced:
      --custom-url string      Custom base URL for the model provider API

Commands:
  account       Manage AI provider accounts (login, logout, usage)
  config        Manage configuration (add, remove models)
  conversation  Manage conversation history [aliases: convo, conv]
                ├─ list    List conversations [alias: ls]
                ├─ print   Print a conversation [aliases: show, view]
                └─ delete  Delete conversations [aliases: rm, remove]
  model         List and inspect models [alias: models]
                ├─ list          List models [alias: ls]
                ├─ info          Show model details
                └─ system-prompt Show rendered system prompt
  mcp           MCP tools
                ├─ list-servers List servers [alias: ls-servers]
                ├─ list-tools   List tools [alias: ls-tools]
                ├─ call-tool    Call a tool
                ├─ info         Server info
                └─ code-desc    Show code mode description
  completion    Generate shell autocompletion scripts (bash, zsh, fish, powershell)

🔧 Troubleshooting

Common Issues

"API key missing: ANTHROPIC_API_KEY not set"

• Ensure the environment variable specified in api_key_env is exported in your shell
• Check for typos in the variable name in both your shell and cpe.yaml
• For OAuth authentication, run cpe account login anthropic instead of using API keys

"no configuration file found"

• Create a cpe.yaml in the current directory or your user config directory (see Configuration File Locations)
• Use --config /path/to/cpe.yaml to specify an explicit path

"model not found: xyz"

• Run cpe model list to see available models
• Check that the ref in your cpe.yaml matches what you're requesting with -m

MCP server fails to start

• Check that the command exists and is executable
• For stdio servers, ensure the command path is correct
• Use cpe --model <model> mcp list-tools <server> to debug tool availability
• Check server logs with cpe --model <model> mcp info <server>

"context length exceeded"

• Use fewer input files or truncate large files
• Check context_window in your model config matches the provider's limits
• Start a new conversation with -n to clear history

Timeout errors

• Increase the selected model profile's timeout or use --timeout 10m
• For MCP servers, adjust the per-server timeout value

Debug Tips

# Check which model is being used
cpe model info <ref>

# View the system prompt being sent (uses -m flag for model selection)
cpe model system-prompt -m <ref>

# Test MCP server connectivity
cpe --model <model> mcp list-tools <server-name>

🤝 Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

1. Fork the repository
2. Create your feature branch (git checkout -b feature/amazing-feature)
3. Commit your changes (git commit -m 'Add amazing feature')
4. Push to the branch (git push origin feature/amazing-feature)
5. Open a Pull Request

See AGENTS.md for detailed development guidelines, code style, and testing practices.

📄 License

This project is licensed under the MIT License.