Companion Connector MCP
Local MCP server enabling ChatGPT to reference large files by path, manage long-running jobs, and register local resources like screenshots and logs without modifying source files.
README
Companion Connector MCP
Local Companion MCP app for ChatGPT New App custom server connection.
V5 capabilities
This version is built for large local work, Fable handoff, screenshots/images, URL reading, media metadata, and future 21-service routing.
Core MCP tools
searchfetchregister_file_pointerread_file_sliceread_result_slicecreate_summary_jobcreate_file_digest_jobcreate_directory_inventory_jobregister_image_pointeringest_image_base64get_image_datacreate_image_inspection_jobingest_chat_transcriptingest_attachment_base64create_fable_bundlerun_fable_bundle
Fable and batch tools
create_question_batchrun_question_batchask_fable_bigcreate_fable_improvement_surveyconnector_health_reportlist_jobslist_fable_runscreate_fable_prompt_fileget_job_statuslist_registered_resources
URL/media/handoff tools
fetch_url_textextract_links_from_urlcreate_url_snapshot_jobregister_video_pointercreate_media_metadata_jobcreate_handoff_queue_itemlist_handoff_queuelist_mcp_servicesdescribe_mcp_service
What it solves
- Big files are referenced by path instead of pasted into chat.
- Large questions can be written into local prompt bundles for Fable.
- Multiple questions can be batched into one file-backed request.
- URLs can be fetched into bounded snapshots.
- Links can be extracted from a page.
- Attachments can be stored as connector resources.
- Screenshots/images can be registered by path or small base64 payload.
- Media/video files can be registered and inspected with ffprobe when available.
- Fable can be given a bundle prompt file with transcript, attachment pointers, full text sections, image metadata, media metadata, and hashes.
- 21 planned MCP service folders are exposed as a service catalog.
- Outputs are written to
results/and read by slice.
Start locally
npm start
Default endpoint:
http://127.0.0.1:8788/mcp
Tests
npm test
npm run test:fable
npm run test:v4
npm run test:v5
Hidden PowerShell 7 startup
.\scripts\start-hidden.ps1
Stop by port:
.\scripts\stop-companion.ps1
Tunnel setup
See TUNNEL_SETUP.md.
Safety boundaries
- Reads are restricted to configured allowed roots.
- Connector writes stay under the connector workspace.
- Source folders are not modified by pointer registration.
_AI_CHATS_ОБЩИЕis not reorganized by this app.- Large outputs are written to
results/and read by slice. - Calls are logged by hash, not full arguments.
- Widget resource uses
text/html;profile=mcp-app.
V6 additions
V6 adds a verified 100-improvement registry and audit layer:
list_implemented_improvementsaudit_100_improvementscreate_support_bundleruntime_metricsvalidate_connector_configcreate_debug_snapshot
Files:
docs/V6_100_IMPLEMENTED_IMPROVEMENTS.mddocs/V6_100_IMPLEMENTED_IMPROVEMENTS.json
Test:
npm run test:v6
V7 no-window startup and 1000 forward improvements
Manual npm start no longer keeps a visible server console open. It starts the connector in the background through PowerShell 7.
Best daily start method with no popup:
wscript.exe .\CompanionConnector-START-HIDDEN.vbs
Or double-click:
CompanionConnector-START-HIDDEN.vbs
Install hidden startup at Windows logon:
.\scripts\install-hidden-task.ps1
Remove hidden startup task:
.\scripts\uninstall-hidden-task.ps1
V7 also adds a 1000-item forward improvement backlog. It is explicitly not falsely marked as implemented:
docs/V7_1000_FORWARD_IMPROVEMENTS.mddocs/V7_1000_FORWARD_IMPROVEMENTS.json
V7 tests:
npm run test:v7
npm run test:windowless
V8 read-only full-folder bridge
V8 adds tools for large read-only folder access without modifying the source folder:
folder_explorercreate_readonly_folder_manifestaudit_readonly_folder_manifestcreate_readonly_folder_content_bundleread_folder_bundle_chunksearch_folder_content_bundlecreate_fable_folder_handoff
Tested target:
J:\ПРОЕКТЫ\G01_All_About_Trading\G01_P09_All_for_TradingView\G01_P09_01_Project\TradingView_INDICATORS\IGOR_ENTER2\research_out
V8 test:
npm run test:v8
The source folder is only read. Manifest, chunks, hashes, and Fable guide are written under CompanionConnector results/.
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
Neon Database
MCP server for interacting with Neon Management API and databases
E2B
Using MCP to run code via e2b.
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.