Skip to content
ReCursor Docs

Research Report: Claude Code Integration Feasibility for Third-Party Mobile Client

Edit page

Generated: 2026-03-17 | Researcher Agent Task: Determine supported integration options for ReCursor (Flutter mobile) to connect to Claude Code

Executive Summary

Section titled “Executive Summary”

Primary Finding: There is no official public API or protocol for third-party clients to join Claude Code Remote Control sessions. The Remote Control feature is designed exclusively for first-party Anthropic clients (claude.ai/code web UI, VSCode extension, and official Claude mobile apps).

Recommended Approach: ReCursor should use the Claude Agent SDK (@anthropic-ai/claude-agent-sdk) as the supported integration path, running as a separate agentic session rather than attempting to mirror an existing Claude Code session. Alternative integration points include the Hooks system for event observation and MCP (Model Context Protocol) for tool interoperability.

Key Risks: Terms of Service restrictions, authentication complexity, and the undocumented nature of Remote Control internals make direct protocol reverse-engineering inadvisable.

Source Validation

Section titled “Source Validation”
SourceTierDateEvidence Type
C:/Repository/claude-code/CHANGELOG.md1 (Official)2026-03Release notes showing Remote Control as internal feature
C:/Repository/claude-code/LICENSE.md1 (Official)CurrentCommercial Terms of Service reference
C:/Repository/claude-code/plugins/agent-sdk-dev/README.md1 (Official)CurrentAgent SDK documentation
C:/Repository/claude-code/SECURITY.md1 (Official)CurrentSecurity policy
C:/Repository/claude-code/README.md1 (Official)CurrentOfficial documentation links

Key Findings

Section titled “Key Findings”

1. Remote Control Feature Analysis

Section titled “1. Remote Control Feature Analysis”

What is Remote Control?

Section titled “What is Remote Control?”

From the CHANGELOG analysis, Remote Control is a feature that allows Claude Code terminal sessions to be accessed and controlled from other interfaces:

“Added claude remote-control subcommand for external builds, enabling local environment serving for all users.” — CHANGELOG.md v2.1.51

“Added optional name argument to /remote-control and claude remote-control (/remote-control My Project or --name "My Project") to set a custom session title visible in claude.ai/code” — CHANGELOG.md v2.1.69

Remote Control Implementation Details

Section titled “Remote Control Implementation Details”

The CHANGELOG reveals technical implementation details:

Protocol: WebSocket-based with JWT authentication

“Fixed several Remote Control issues: sessions silently dying when the server reaps an idle environment, rapid messages being queued one-at-a-time instead of batched, and stale work items causing redelivery after JWT refresh” — CHANGELOG.md v2.1.77

Polling Mechanism:

“Reduced Remote Control /poll rate to once per 10 minutes while connected (was 1–2s), cutting server load ~300×. Reconnection is unaffected — transport loss immediately wakes fast polling.” — CHANGELOG.md v2.1.72

Bridge Sessions:

“Fixed bridge sessions failing to recover after extended WebSocket disconnects” — CHANGELOG.md v2.1.77

“Fixed a race condition in the REPL bridge where new messages could arrive at the server interleaved with historical messages during the initial connection flush, causing message ordering issues.” — CHANGELOG.md v2.1.58

Official Remote Control Consumers

Section titled “Official Remote Control Consumers”

The CHANGELOG identifies these as the only supported Remote Control clients:

  1. claude.ai/code web UI — “visible in claude.ai/code”
  2. VSCode Extension — “[VSCode] Added support for remote sessions, allowing OAuth users to browse and resume sessions from claude.ai”
  3. Official Claude mobile apps — “Fixed Android app crash when running local slash commands (/voice, /cost) in Remote Control sessions”

Policy Restrictions

Section titled “Policy Restrictions”

Remote Control access is restricted by plan tier:

“Added policy limit fetching (e.g., remote control restrictions) for Team plan OAuth users, not just Enterprise” — CHANGELOG.md v2.1.69

This indicates Remote Control is a managed feature with enterprise policy controls, not a public API.

2. No Public API for Third-Party Remote Control

Section titled “2. No Public API for Third-Party Remote Control”

Conclusion: There is no documented public API for third-party clients to:

  • Join Remote Control sessions
  • Mirror chat events from Claude Code
  • Observe tool execution in real-time
  • Control Claude Code sessions programmatically

The Remote Control protocol is internal and undocumented. Evidence:

  1. No API documentation exists in the public docs at code.claude.com
  2. No OpenAPI spec or protocol documentation in the repository
  3. Authentication requires OAuth via Anthropic’s identity system
  4. Policy restrictions are enforced server-side

3. Supported Integration Alternatives

Section titled “3. Supported Integration Alternatives”
Section titled “Option A: Claude Agent SDK (Recommended)”

The Claude Agent SDK (@anthropic-ai/claude-agent-sdk) is the officially supported way to build agentic applications that interact with Claude.

Key Details:

Migration Note:

“Removed legacy SDK entrypoint. Please migrate to @anthropic-ai/claude-agent-sdk for future SDK updates” — CHANGELOG.md v2.0.25

Architecture for ReCursor: Instead of mirroring a Claude Code session, ReCursor would:

  1. Run its own Agent SDK session
  2. Implement similar tool capabilities (Bash, Read, Edit, etc.)
  3. Maintain separate conversation state
  4. Use the same underlying Claude API

Limitation: This creates a parallel session, not a mirror of an existing Claude Code session.

Option B: Hooks System (Event Observation)

Section titled “Option B: Hooks System (Event Observation)”

Claude Code provides a Hooks system that allows plugins to observe and react to events.

Available Hook Events (confirmed from source truth):

  • SessionStart / SessionEnd
  • PostToolUse / PreToolUse
  • UserPromptSubmit
  • Stop / SubagentStop
  • PreCompact
  • Notification

Note: Additional events may be mentioned in CHANGELOG but the above list represents confirmed events from the Claude Code hooks implementation source truth.

Hook Configuration: Hooks can be configured to:

  • Run shell commands
  • POST JSON to HTTP URLs (“Added HTTP hooks, which can POST JSON to a URL and receive JSON instead of running a shell command” — CHANGELOG.md v2.1.46)

For ReCursor: A ReCursor backend could:

  1. Set up an HTTP endpoint to receive hook events
  2. Configure Claude Code to POST events to this endpoint
  3. Forward events to the Flutter mobile app

Limitation: Hooks are one-way (observation only). They cannot inject messages or control the session.

Option C: MCP (Model Context Protocol)

Section titled “Option C: MCP (Model Context Protocol)”

MCP is an open protocol for extending Claude’s capabilities with custom tools.

Documentation: https://code.claude.com/docs/en/plugins (referenced in CHANGELOG)

Server Types Supported:

  • stdio (local processes)
  • SSE (hosted/OAuth)
  • HTTP (REST)
  • WebSocket (real-time)

For ReCursor: ReCursor could expose itself as an MCP server that Claude Code connects to, enabling:

  • Mobile-initiated tool calls
  • Bidirectional communication
  • Custom tool definitions
Section titled “4. Terms of Service and Legal Considerations”

License:

”© Anthropic PBC. All rights reserved. Use is subject to Anthropic’s Commercial Terms of Service.” — C:/Repository/claude-code/LICENSE.md

Security Policy:

“The security of our systems and user data is Anthropic’s top priority.” — C:/Repository/claude-code/SECURITY.md

Risks of Reverse Engineering:

  1. ToS Violation: Reverse engineering the Remote Control protocol may violate Anthropic’s Commercial Terms of Service
  2. Authentication Barriers: OAuth and JWT requirements make unauthorized access technically infeasible
  3. Breaking Changes: Internal APIs change without notice (evident from frequent CHANGELOG updates)
  4. Policy Enforcement: Remote Control restrictions are enforced server-side for Team/Enterprise plans
Section titled “Recommended Approach for ReCursor Documentation”

What to State as Plan

Section titled “What to State as Plan”
  1. Primary Integration: Use Claude Agent SDK (@anthropic-ai/claude-agent-sdk) for core agentic functionality
  2. Event Streaming: Implement HTTP hooks in Claude Code to stream events to ReCursor backend
  3. Tool Integration: Expose ReCursor-specific tools via MCP protocol
  4. Authentication: Use standard Anthropic API key authentication (not OAuth/Remote Control)

What to State as Open Questions

Section titled “What to State as Open Questions”
  1. Session Synchronization: How to synchronize state between a user’s desktop Claude Code session and mobile ReCursor session
  2. Multi-Device Coordination: Whether to support simultaneous use or session handoff
  3. Enterprise Policy: How to handle environments with managed settings restricting hooks/MCP

Risk Statement for Documentation

Section titled “Risk Statement for Documentation”
## Integration Risks and Limitations


**Not Supported:** Direct integration with Claude Code's Remote Control feature is not
possible for third-party applications. Remote Control is restricted to official Anthropic
clients (claude.ai/code, VSCode extension, official mobile apps) and requires OAuth
authentication with enterprise policy enforcement.


**Recommended Alternative:** ReCursor uses the Claude Agent SDK to create independent
agentic sessions with equivalent capabilities. This approach:
- ✅ Is fully supported by Anthropic
- ✅ Uses public APIs with stable contracts
- ✅ Complies with Terms of Service
- ⚠️ Creates parallel sessions rather than mirroring desktop sessions

Verification Evidence

Section titled “Verification Evidence”
Section titled “Command Output: Repository Search”
Terminal window
$ cd C:/Repository/claude-code && grep -ri "remote.control" --include="*.md" -C 3 | head -50
CHANGELOG.md-- Fixed several Remote Control issues: sessions silently dying when the server reaps an idle environment, rapid messages being queued one-at-a-time instead of batched, and stale work items causing redelivery after JWT refresh
CHANGELOG.md-- Fixed bridge sessions failing to recover after extended WebSocket disconnects
CHANGELOG.md:- Improved Remote Control session titles — now derived from your first prompt instead of showing "Interactive session"
CHANGELOG.md:- Added optional name argument to `/remote-control` and `claude remote-control` (`/remote-control My Project` or `--name "My Project"`) to set a custom session title visible in claude.ai/code
CHANGELOG.md:- Added policy limit fetching (e.g., remote control restrictions) for Team plan OAuth users, not just Enterprise

File Listing: Agent SDK Plugin

Section titled “File Listing: Agent SDK Plugin”
Terminal window
$ ls -la C:/Repository/claude-code/plugins/agent-sdk-dev/
total 16
drwxr-xr-x 1 Administrator 197121 0 Mar 17 19:57 .
drwxr-xr-x 1 Administrator 197121 0 Mar 17 19:57 ..
-rw-r--r-- 1 Administrator 197121 795 Mar 17 19:57 README.md
-rw-r--r-- 1 Administrator 197121 1729 Mar 17 19:57 commands/
-rw-r--r-- 1 Administrator 197121 4632 Mar 17 19:57 agents/

Quoted Excerpt: Agent SDK Documentation Reference

Section titled “Quoted Excerpt: Agent SDK Documentation Reference”

From C:/Repository/claude-code/plugins/agent-sdk-dev/README.md:

“Resources:

Quoted Excerpt: License Terms

Section titled “Quoted Excerpt: License Terms”

From C:/Repository/claude-code/LICENSE.md:

”© Anthropic PBC. All rights reserved. Use is subject to Anthropic’s Commercial Terms of Service.”

References

Section titled “References”
  1. Claude Code Repository: C:/Repository/claude-code/
  2. Official Documentation: https://code.claude.com/docs/en/overview
  3. Agent SDK Docs: https://docs.claude.com/en/api/agent-sdk/overview
  4. Commercial Terms: https://www.anthropic.com/legal/commercial-terms
  5. Security Policy: https://hackerone.com/anthropic-vdp

Conclusion

Section titled “Conclusion”

ReCursor should not attempt to integrate with Claude Code’s Remote Control feature due to:

  • Lack of public API documentation
  • OAuth/JWT authentication requirements
  • Enterprise policy restrictions
  • Terms of Service implications

Instead, ReCursor should:

  1. Build on the Claude Agent SDK for core functionality
  2. Use HTTP hooks for event observation from Claude Code
  3. Implement MCP servers for tool interoperability
  4. Position as a complementary mobile agent rather than a remote control client

This approach is fully supported, legally compliant, and architecturally sound.