Agent access

Your repositories. One agent workflow.

Give your coding agent one skill to prepare your portfolio, gather relevant project history, upload context, run checks and submit the repositories you choose.

Download the intake skill Workflow manifest API schema

MCP or HTTP for service actions. Local Git for code changes. Computer use only when an export or owner approval needs it.

Start with this instruction

Read https://codemerchant.ai/skills/codemerchant-intake/SKILL.md and its linked references. Sign in with my GitHub login, prepare my selected repositories for CodeMerchant, collect relevant context and local agent traces from Claude, Codex and other coding tools within the company/repository scope I authorize, review and upload those files to the correct destinations, fix reviewable issues, and verify each submission. Ask me for any missing access or owner decisions.

Save SKILL.md in your agent’s skill directory. Download its linked references and upload helper alongside it, or let your agent retrieve them from this site.

API & file transfers External context guide Repository preparation Resumable upload helper

Repository copying skill

Bring your issues and pull requests along

Copy a repository to your own account with its commits, issues, pull requests, comments and closing links. This skill guides your agent through a complete inventory, faithful copying and verification, with original authors and dates attributed where GitHub cannot preserve them.

Read https://codemerchant.ai/skills/copy-github-collaboration-history/SKILL.md and its linked API procedure. Help me choose the source and destination repositories, then copy their Git and collaboration history and verify the result.

Download the copying skill API procedure Inventory helper Agent configuration

Give your agent the instruction above, or save these files together with their folder structure. The inventory helper needs Python 3 and an authenticated GitHub CLI. No CodeMerchant account is required.

Choose the portfolio

You authorize the repositories and sources. Your agent inventories connected private GitHub repositories and records anything it cannot access.

Bring the project history

Your agent gathers relevant Linear issues, Slack threads, specs and coding sessions through authorized connectors, or helps you export them.

Prepare, upload and check

Your agent reviews and formats context, attaches it to the right repositories, verifies extraction and runs quality and privacy checks.

Fix and submit

Your agent prepares reviewable code fixes, rescans with context and submits your authorized selection. You retain history-publication and commercial decisions.

Bring the work behind the code

Connect the sources your agent can read, or export the relevant project yourself. Useful context includes Linear issues and comments, Slack project threads, design documents, decisions and local agent traces from Claude, Codex and other coding tools. Your agent collects matching sessions from your computer, preserves prompts, replies and tool activity, and uploads reviewed copies for each authorized company and repository.

Choose which sources may be shared. The original uploaded files are saved for administrator review. Your agent should review private material first, keep source references, verify each attachment and flag missing exports.

See what to export from each source →

Your agent handles

Inventory and a result for every selected repository.
Relevant context collection, formatting and verified uploads.
Checks, reviewable fixes, rescans and authorized submission.

You control

GitHub access, repository selection and context sharing.
Data License acceptance and submission authorization.
Publishing rewritten history, merging changes and commercial decisions.

Connect your agent

First connect GitHub and accept the Data License in your dashboard. Then create a 90-day token. Enable uploads and submission for the full intake workflow; existing analysis tokens keep their current permissions.

Store the token in your secret manager as CODEMERCHANT_API_TOKEN. It appears once and is stored only as a hash. Expand the variable using your client’s supported configuration; never paste a token into chat or a repository.

{
  "mcpServers": {
    "codemerchant": {
      "type": "http",
      "url": "https://codemerchant.ai/api/mcp",
      "headers": {
        "Authorization": "Bearer ${CODEMERCHANT_API_TOKEN}"
      }
    }
  }
}

Tools for the whole intake

get_repository_report

Read the latest source-free report and scan status for a connected repository.

GET /api/v1/reports/{owner}/{name}

scan_repositories

Run an owner-selected bounded batch using saved completed context applicable to each repository. Inspect scanned, failed and remaining. On data_license_required, follow the returned message: show the owner the terms and accept only after their explicit yes.

POST /api/v1/scans

get_data_license

Read the current Data License version, summary, acknowledgements, agreement link and whether the owner has accepted it.

GET /api/v1/data-license

accept_data_license

Accept the Data License for the owner. First show the owner the summary, agreement link and every acknowledgement word for word from get_data_license and ask whether they accept. Call this only after they explicitly say yes, passing that version, every acknowledgement true, and their reply quoted as ownerConfirmation. Never accept without their explicit yes.

POST /api/v1/data-license

list_repositories

Read the saved connected repository inventory.

GET /api/v1/repositories

refresh_inventory

Refresh connected repositories and report unavailable installations.

POST /api/v1/repositories

context_upload

Initiate, complete, check, list or change the scope of uploads. Transfer file bytes to the returned uploadUrl using HTTP PUT.

POST /api/v1/context-uploads

scan_with_context

Scan one selected repository. Omit contextIds or send an empty array to reuse all saved completed context applicable to it; nonempty IDs select specific uploads. Use refresh true after reviewed fixes. Requires owner Data License acceptance.

POST /api/v1/context-scans

list_submissions

Read saved submission IDs and review stages.

GET /api/v1/submissions

submit_repositories

Submit the owner-authorized repository selection with completed analyses. Requires owner Data License acceptance. Does not approve a sale or payout. A saved response means the selection is stored; listing "unavailable" means only the read-back failed, so re-read list_submissions instead of resubmitting. On github_rate_limited, wait retryAfterSeconds before retrying.

POST /api/v1/submissions

refresh_submission

Refresh the analysis of one existing submission after fixes.

POST /api/v1/submissions/refresh

History cleanup is a separate, reviewable step

A normal code fix leaves earlier commits intact. Your agent can prepare and validate a rewrite in a private mirror, preserve provenance and show you the affected refs before publication. Publishing rewritten history requires your explicit approval.

Read the history-cleanup procedure →

Repository ingestion currently uses connected private GitHub repositories. Standalone source archives and final vendor-package delivery are not supported. Website remediation can require owner GitHub approval; your agent can also prepare fixes through its own authorized Git tools. Submission records your application for review, not a sale or payout.