Anchor

Anchor is a local-first, MCP-compatible repo and org memory tool for GitHub PR history, codebase indexing, tests, regressions, architecture memory, cross-repo impact, and AI coding agent context.

It gives AI coding agents concise, sanitized, evidence-backed context before code edits without SaaS, CLI telemetry, GitHub write access, or remote LLM calls.

Install:

npm install -g @pratik7368patil/anchor

Try the offline demo before setting up GitHub auth:

npx @pratik7368patil/anchor demo

Anchor has no CLI telemetry. If it helps, public proof comes from voluntary signals like GitHub stars, forks, issues, and aggregate npm/GitHub traffic.

Use it inside a GitHub-backed repo:

gh auth login
anchor init
anchor init --target cursor,codex,claude-code
anchor init --all-targets
anchor init --target cursor --no-autosync
anchor demo
anchor prompts --target codex
anchor index
anchor index-all --concurrency 6
anchor index-code
anchor context "Refactor auth cache" --file src/auth/cache.ts --strict
anchor explain src/auth/cache.ts
anchor explain src/auth/cache.ts --share
anchor architecture
anchor architecture --file src/auth/cache.ts
anchor architecture --area api
anchor architecture --check
anchor review
anchor review --share
anchor health
anchor org init --org my-org
anchor org add-repo --org my-org
anchor org add-repo --org my-org --search api
anchor org add-repo my-org/backend-api --org my-org --group backend
anchor org sync --org my-org
anchor org graph --org my-org
anchor org graph --org my-org --open
anchor org map --org my-org --open
anchor org impact --org my-org --repo my-org/backend-api --strict --open
anchor org ci --org my-org --strict --min-coverage 70 --html
anchor rules init
anchor rules validate
anchor rules suggest
anchor rules add --id api-contract --category api_contract --text "Keep API stable." --pr-number 123 --pr-url https://github.com/owner/repo/pull/123
anchor rules check-evidence
anchor doctor

Which command should I run?

First-time setup:

anchor init
anchor index --limit 200
anchor doctor

Current-code context only, no GitHub auth:

anchor index-code
anchor architecture
anchor health

Full PR history:

anchor index-all --concurrency 2

Daily refresh is installed automatically by anchor init. To refresh manually:

anchor sync

Before AI agent edits:

anchor plan "Add API integration" --file src/api/routes.ts
anchor test-command src/api/routes.ts
anchor explain src/api/routes.ts

Before API, auth, access, schema, SDK, shared-package, or cross-repo work:

anchor org sync --org my-org --no-graph --concurrency 2
anchor org graph --org my-org
anchor org impact --org my-org --repo my-org/backend-api --strict --open

Command-specific options

anchor index: Use --repo owner/name when git remote detection is unavailable, --limit 50 for a fast first pass, --all for full history through the normal command, --since YYYY-MM-DD for targeted backfill, --force to rebuild local derived records, --no-code to skip code indexing, and --concurrency 1-10 to tune patch enrichment pressure.

anchor index-all: Use for complete merged PR history. Prefer --concurrency 1 or --concurrency 2 on large repos. Use --no-code when code context is already fresh.

anchor sync: Use after the first index. It is incremental and safe to rerun. Scheduled autosync runs this daily with --all --concurrency 2. Add --all for a full catch-up from the sync cursor, --since YYYY-MM-DD for a specific window, --no-code for PR-only refresh, and --concurrency 1-10 when tuning rate-limit pressure.

anchor index-code: Use when you only need current-code context or do not have GitHub auth. Add --force when anchor health reports stale code records.

Anchor automatically chooses progress output: modern live progress in interactive terminals, plain line logs in CI/non-TTY shells, and no progress for JSON output. Org commands show the active repo, phase, counts, elapsed time, and last update age across GitHub fetches, PR SQLite indexing, code indexing, architecture indexing, and graph creation.

anchor plan: Use --file path for a likely target file, --symbol name for a likely contract or implementation point, --strict for high-risk work, and --json for automation.

anchor explain and anchor review: Use --share for Slack or PR-comment Markdown. Use --diff-file change.diff on review when checking a saved diff. Use --strict for risky diffs and --json for tooling.

anchor architecture: Use --file path for file-level guidance, --area api for one architecture area, --check for the current diff, --diff-file change.diff for saved diffs, --map --format mermaid for docs, --map --format json for tooling, and --write-doc only when you intentionally want ANCHOR_ARCHITECTURE.md.

anchor org ...: Use --org my-org on every org command. Autosync runs daily org sync --no-graph --concurrency 1 and weekly org graph for org configs that already exist when anchor init runs. Run anchor org add-repo --org my-org to search and multi-select readable GitHub repos, add --search text to prefill the picker, and use --include-archived only when archived repos are intentionally needed. Use the explicit anchor org add-repo owner/name --org my-org form for scripts/non-TTY shells, --group to classify selected repos, and --alias only for one explicit repo. Use --repo to retry one repo, --code-only or --prs-only with org index, --no-graph with org index/org sync to postpone cross-repo graph rebuilds, org graph to rebuild only edges/API consumers, org graph --open to inspect the graph in a local browser UI, --concurrency 1-3 with org clone and org sync, --diff-file and --strict with org impact, --min-coverage with org ci, and --html/--open/--output with org map, org impact, and org ci to generate local human-readable HTML reports.

Then reload your selected AI tool and use the MCP tools anchor_get_context, anchor_explain_file, anchor_review_diff, anchor_get_architecture, anchor_check_architecture, and anchor_check_cross_repo_impact. If a tool cannot call MCP yet, use anchor context "<task>" --file path --strict and paste the sanitized Markdown into the agent.

Existing PR indexing commands use GitHub GraphQL first for batched PR metadata, comments, reviews, commits, labels, and changed files. Anchor uses REST only to enrich PR file patches, caps GraphQL page size below GitHub's nested node ceiling, adapts GraphQL page size from live rate-limit cost, and saves a local resume checkpoint for full-history runs, so anchor index, anchor index-all, and anchor sync are more efficient without adding another command. Transient GraphQL network/HTML gateway failures retry before Anchor falls back or fails clearly; GraphQL rate/resource limits reduce page size or defer instead of using the older REST PR-detail crawler.

Use anchor_get_context with strict: true when the agent should only receive non-stale, high-confidence evidence.

Anchor indexes PR history, local code chunks, likely related tests, regression memory, architecture patterns, and team-approved rules. anchor health and anchor_index_status include a local coverage score and autosync status. All data stays in .anchor/index.sqlite on your machine.

Org Memory is opt-in. anchor org ... commands store allowlisted repo clones and one org SQLite database under ~/.anchor/orgs/<org>/, then expose cross-repo context through anchor_get_org_context, anchor_check_cross_repo_impact, anchor_find_api_consumers, anchor_get_org_architecture, and anchor_org_index_status.

Cross-repo edges and API consumers are created during the org graph phase. If a large anchor org sync is taking too long after repo indexing completes, run anchor org status --org my-org in another terminal to see the heartbeat, then split future runs with anchor org sync --org my-org --no-graph and anchor org graph --org my-org --open. If a recent sync is interrupted after PR/code indexing but before graph completion, rerunning anchor org sync resumes graph work and skips redundant PR fetches for repos that already completed PR sync.

Architecture Memory is refreshed by anchor index, anchor index-all, anchor sync, and anchor index-code. It gives AI coding agents deterministic current-code guidance about file areas, import direction, symbols, repeated folder patterns, and nearby test conventions before adding APIs, services, components, hooks, tests, or refactors.

anchor demo runs offline with bundled fixtures and sample code. --share on explain and review produces compact Markdown for Slack or PR comments.

Docs, adoption signals, and feedback links: https://anchor-mcp.netlify.app

anchor init also adds .anchor/ to .git/info/exclude, keeping the local SQLite index out of git without changing .gitignore. It installs local autosync by default using your OS scheduler and logs to ~/.anchor/logs/autosync/. Use anchor init --no-autosync or anchor init --autosync off to opt out.

Full documentation: https://github.com/pratik7368patil/anchor#readme