"Description": "Capability matrix for ABP Studio AI Agent modes, built-in tools, Studio automation tools, subagents, MCP tools, and hidden tool boundaries."
"Description": "Capability matrix for ABP Studio AI Agent modes, built-in tools, Studio automation tools, subagents, and hidden tool boundaries."
}
```
@ -17,7 +17,7 @@
}
````
ABP Studio exposes tools to the AI Agent based on the selected mode, run context, tool settings, permissions, and connected MCP servers. The tool list is locked per running Agent-mode session to keep prompt caching and execution stable while other sessions run in parallel.
ABP Studio exposes tools to the AI Agent based on the selected mode, run context, tool settings, and permissions. The tool list is locked per running Agent-mode session to keep prompt caching and execution stable while other sessions run in parallel.
## Mode Capability Matrix
@ -41,7 +41,6 @@ ABP Studio exposes tools to the AI Agent based on the selected mode, run context
| Update active plan steps | Yes | No | No |
| Search/read ABP docs directly | No | No | Yes |
| Studio automation tools | Yes | No | No |
| Connected MCP tools | Yes | No | No |
| Spawn subagents | Yes | Yes | Yes |
ABP documentation tools are available to Ask mode and ABP documentation subagents. The main Agent-mode tool list does not include direct ABP docs search/read tools; Agent mode can use the ABP documentation subagent when documentation research is needed.
@ -138,16 +137,6 @@ The `spawn_subagent` tool delegates bounded research work to read-only specialis
Subagents are read-only. They are intended for parallel research and return summarized findings to the main agent.
## MCP Tools
Connected MCP server tools are added to Agent mode when:
- The MCP server is configured and enabled.
- The server connection is successful.
- The individual MCP tool is not disabled in settings.
MCP resources are visible in settings and can be opened for inspection, but the Agent-mode callable surface is composed from MCP tools.
## Tool Execution Gates
ABP Studio serializes selected operations to reduce parallel-session conflicts.
"Description": "Configuration reference for ABP Studio AI Agent models, reasoning, context limits, parallel sessions, permissions, AI rules, learned lessons, and MCP servers."
"Description": "Configuration reference for ABP Studio AI Agent models, reasoning, context limits, parallel sessions, permissions, AI rules, and learned lessons."
}
```
@ -17,7 +17,7 @@
}
````
AI Agent configuration controls model selection, execution limits, enabled tools, tool permissions, AI rules, learned lessons, and Model Context Protocol (MCP) servers.
AI Agent configuration controls model selection, execution limits, enabled tools, tool permissions, AI rules, and learned lessons.
## Model Provider
@ -136,43 +136,3 @@ Always-apply rules are injected into the agent's system prompt. Non-always-apply
The `save_lesson` tool records verified corrections into `.abpstudio/ai-rules/ai-learned-lessons.mdc`. Learned lessons are injected as high-priority context in future sessions.
The agent is instructed to save a lesson only after a user correction is verified against source or tool output. When the learned lessons file becomes large, Studio can consolidate it with the text processor model.
## MCP Servers
ABP Studio can connect to external MCP servers and expose their tools to Agent mode. MCP servers can be configured with:
| Transport | Configuration |
| --- | --- |
| Stdio | Command, arguments, and environment variables. |
| HTTP | URL and headers. |
Studio imports MCP configuration from Cursor, Claude, VS Code, Windsurf, and bare MCP server JSON formats. Studio exports MCP server configuration in the standard `mcpServers` JSON shape.
Connected MCP servers show their connection status, tool count, tools, and resources. Individual MCP tools can be disabled. Disabled MCP tools are omitted from Agent mode. MCP resources can be opened from settings for inspection.
MCP tools are added only for connected and enabled servers. Plan and Ask modes do not receive MCP tools.
The ABP CLI includes `abp mcp-studio`, a stdio bridge for external AI tools. The command forwards JSON-RPC MCP messages to ABP Studio's local MCP endpoint and requires ABP Studio to be running.
@ -72,7 +72,7 @@ Blocking review content includes user notes and AI suggestions with severities o
## Sending Review Findings To The Agent
The Git panel can send review findings to ABP Agent. Studio builds a prompt that includes:
The Git panel can send review findings to AI Agent. Studio builds a prompt that includes:
- A request to review and apply the notes for the current Git changes.
- File paths.
@ -93,13 +93,13 @@ The generator returns a concise past-tense commit message. It is intended to sum
## Merge Conflict Handoff
When unresolved merge conflicts exist, the Git panel can send the conflict list to ABP Agent. The generated prompt asks the agent to resolve conflicts and request clarification when the correct resolution is ambiguous.
When unresolved merge conflicts exist, the Git panel can send the conflict list to AI Agent. The generated prompt asks the agent to resolve conflicts and request clarification when the correct resolution is ambiguous.
Conflict handoff can use the current session or create a new session. It starts with analysis skipped so the prompt is centered on the conflict files.
## GitHub Issue Handoff
GitHub issue details can be sent to ABP Agent from Studio when GitHub integration is connected.
GitHub issue details can be sent to AI Agent from Studio when GitHub integration is connected.
The generated issue prompt includes:
@ -115,7 +115,7 @@ Comments can be excluded from the AI prompt. Images that belong to excluded comm
## GitHub Pull Request Handoff
Pull request details can be sent to ABP Agent when the current branch matches the PR branch.
Pull request details can be sent to AI Agent when the current branch matches the PR branch.
The generated PR prompt includes:
@ -143,4 +143,4 @@ Image handling still depends on the selected model's image support. If the model
Agent URL fetching can enrich GitHub URLs with issue or pull request information when GitHub integration is available. Pull request URL context can include review comments up to Studio's fetch limits.
This is separate from the explicit GitHub issue/PR "Send to ABP Agent" actions in the Git UI.
This is separate from the explicit GitHub issue/PR "Send to AI Agent" actions in the Git UI.
@ -29,7 +29,7 @@ The selected mode determines the agent's effective permission boundary.
| Mode | Purpose | Tool boundary |
| --- | --- | --- |
| Agent | Coding and execution | Can read and write files, run shell commands, add migrations, run enabled Studio tools, update active plan steps, and call connected MCP tools. |
| Agent | Coding and execution | Can read and write files, run shell commands, add migrations, run enabled Studio tools, and update active plan steps. |
| Plan | Read-only planning | Can inspect code and documentation, create or update plan files, and use read-only subagents. It cannot modify solution files or run shell/write tools. |
| Ask | Read-only question answering | Can inspect code, search/read ABP documentation, fetch relevant AI skills, and use read-only subagents. It cannot create plans or modify files. |
@ -94,4 +94,4 @@ When a run profile is active, the system prompt includes information about runna
## Privacy Boundary
The agent uses the files, prompts, tool outputs, attachments, URLs, and connected tool results supplied to a session. Sensitive files can be excluded with `.abpignore`; tool permissions restrict shell commands, domain fetches, and downloads.
The agent uses the files, prompts, Studio tool outputs, attachments, and URLs supplied to a session. Sensitive files can be excluded with `.abpignore`; tool permissions restrict shell commands, domain fetches, and downloads.
"Description": "Technical guidance for coding with ABP Studio AI Agent, including mode selection, scopes, prompting, verification, Git review handoff, MCP tools, and parallel sessions."
"Description": "Technical guidance for coding with ABP Studio AI Agent, including mode selection, scopes, prompting, verification, Git review handoff, Studio tools, and parallel sessions."
}
```
@ -25,7 +25,7 @@ Use the mode that matches the expected side effect.
@ -62,7 +62,7 @@ This keeps the prompt aligned with the actual Git state and avoids missing files
## AI Review Handoff
AI Review is a review workflow, not an implementation workflow. After suggestions are produced, send the review content to ABP Agent when the intent is to apply fixes.
AI Review is a review workflow, not an implementation workflow. After suggestions are produced, send the review content to AI Agent when the intent is to apply fixes.
The handoff prompt contains verified suggestions, user notes, file paths, and line references. The agent then runs in the normal coding workflow and can edit files, run builds, and update active plan steps depending on the session mode and tools.
@ -106,12 +106,6 @@ Workflow actions can configure:
- C# target package.
- Whether C# contract classes are generated.
## MCP Tools
Use MCP tools for external systems that are not built into ABP Studio. MCP server tools are available only in Agent mode and only when the server is connected and the tool is enabled.
Because MCP tools can have arbitrary side effects depending on the server, keep MCP servers limited to trusted tools and disable tools that should not be callable by the agent.
## AI Rules And Lessons
Use always-apply AI rules for standards that should affect every run, such as architecture rules, naming conventions, security requirements, or team-specific ABP patterns.
@ -134,6 +128,6 @@ Common boundaries:
- Files matched by `.abpignore` are inaccessible.
- Image attachments require a model with image support.
- Shell, URL fetch, and download operations can wait for permission.
- Disabled Studio tools and disabled MCP tools are omitted from the tool list.
- Disabled Studio tools are omitted from the tool list.
- Plan and Ask modes cannot modify files.
- Hidden internal Studio tools are not exposed to the agent.
@ -101,7 +101,7 @@ This pane is dedicated to managing Kubernetes services. It simplifies the proces
### AI Agent
The AI Agent is an integrated coding agent in ABP Studio. It can answer ABP-related questions, create implementation plans, and work on the current solution with controlled access to files, Studio tools, workflows, Git context, and configured MCP tools.
The AI Agent is an integrated coding agent in ABP Studio. It can answer ABP-related questions, create implementation plans, and work on the current solution with controlled access to files, Studio tools, workflows, and Git context.
enisn
4 months ago
