AI coding tools comparison

Multica does not replace the AI coding tool you use: a runtime invokes the tool already installed and signed in on the computer, then writes progress and results back to the issue. If you already use one of these tools, you can usually keep it. If you have not picked one yet, weigh four things:

• whether it is already installed on the computer;
• which models your account can use;
• whether Multica needs to configure MCP for it;
• which skills the team wants to reuse.

Install steps are in Install AI coding tools.

Supported toolsCopy Anchor Link

Try itFilter AI coding tools by what you needReplay

I need

Session resumptionMultica-managed MCPModel picked in Multica

ToolsMatching tools: 28 of 28

• Antigravity
• Claude Code
• CodeBuddy
• CodeArts
• Codex
• GitHub Copilot CLI
• Cursor Agent
• DeepSeek Harness
• DevEco Code
• Grok
• Hermes
• Kimi CLI
• Kiro CLI
• MiniMax Code
• Muse Code
• Devin CLI
• OpenClaw
• OpenCode
• Pi
• Oh-My-Pi
• Qoder CLI
• Qoder CN CLI
• Qwen Code
• QwenPaw
• Reasonix
• Trae CLI
• Dim
• ZeroClaw

"Multica-managed MCP" means you can define MCP servers in the agent configuration and Multica passes them to the tool before a run. The column only says whether Multica can configure it for you, not whether the tool has other ways to configure MCP.

Tools with — under "Model picked in Multica" take their model from their own configuration; see Model sources.

Devin supports stdio MCP servers; HTTP/SSE entries are passed only if the CLI advertises support. Devin skills are imported from $XDG_CONFIG_HOME/devin/skills (default ~/.config/devin/skills) or %APPDATA%\devin\skills on Windows, then the legacy cognition/skills root, then ~/.codeium/<channel>/skills (windsurf, windsurf-next, windsurf-insiders, in that order), and finally ~/.agents/skills. On duplicate skill keys, the first root wins for both listing and import.

Model sourcesCopy Anchor Link

When you create an agent, the model list comes from the corresponding runtime. Some tools expose a fixed set of model names; others return available models based on local configuration, the signed-in account, and subscription entitlements.

If you do not pick a model, Multica uses the tool's own default. If the list is empty, confirm the runtime is online and the tool is signed in, then refresh the model list.

Devin's catalog can include models unavailable to your account. Choose a canonical model ID, an unambiguous session model name or catalog family slug/alias (such as opus), or leave the model empty for the CLI default. Multica confirms the actual session model before sending the prompt; a fallback to another family is rejected. Ambiguous names, arbitrary partial names, or unavailable variants require a canonical ID rather than a guessed match. Usage is attributed to the reported model, not the launch flag. Adaptive usage stays under the reported Adaptive routing ID, not an inferred underlying model.

Devin reads WINDSURF_API_KEY when set, otherwise the login saved by devin auth login; unknown ACP auth methods do not block that stored login. Multica preserves Devin's configured session mode rather than raising permissions. If it defaults to plan, switch to an editing mode in Devin before dispatching work that must change files.

QwenPaw, MiniMax Code, and ZeroClaw are exceptions: they do not accept a model chosen from Multica, so the picker is inert and shows "Managed by runtime". Choose the model in the tool's own configuration instead. An empty model list for these tools is expected and does not mean the runtime is offline.

DeepSeek Harness advertises its model catalog from the runtime itself (dsh --profile multica --list-models). Its model ids use the provider/model form (for example deepseek-official/deepseek-chat), so select the full id from the list.

Session resumptionCopy Anchor Link

Tools marked ✓ can continue an existing session in a later run, provided the original session still exists and the run is executed by a runtime that can reach it. MiniMax Code 0.1.2 advertises no session-loading capability over ACP, so Multica starts a fresh MCode session for a later run.

Multica stores the session identifier the tool returns. The next time work continues on the same issue, the runtime tries to resume that session; if it cannot, it starts a new session and keeps going. Pi stores a local session file path, so it depends more on the original computer and file.

Retry and session-resumption rules are in Runs.

MCP configurationCopy Anchor Link

Antigravity, GitHub Copilot CLI, DevEco Code, Muse Code, and ZeroClaw do not read the MCP configuration on an agent yet, so the field is not shown when creating agents for them.

Managed MCP delivery for Pi requires Pi 0.99.0 or later. Older Pi versions continue to run agents that have no MCP configuration.

Servers already configured on the runtime's machine are merged with the agent's configuration on that machine; see Local MCP configuration.

Skill injectionCopy Anchor Link

Before a run starts, Multica writes the skills bound to the agent into the paths in the table above, where each tool discovers them through its own native mechanism.

Skills that already exist in the repository are never overwritten. On a name collision, Multica uses a different directory name for the injected copy; after the run, it cleans up only the files it created.

Codex, Hermes, and OpenClaw are handled slightly differently:

• Codex uses a per-run CODEX_HOME, so skills are never written into the system-level Codex directory.
• Hermes uses a per-run HERMES_HOME. Existing local configuration is still readable, but bound skills and session data are isolated per run. This isolation kicks in only when the agent has skills bound; runs without bound skills use the machine's original Hermes home.
• OpenClaw uses skills/ under the current run's working directory, and the per-run configuration sets that directory as the workspace.

For creating and binding skills, see Skills.

Next stepsCopy Anchor Link

• Create and configure an agent — model and thinking-effort settings.
• Install AI coding tools — install and detection steps.
• Daemon and runtimes — runtime registration and concurrency.