bd-explore

Ask a beads store questions, the way codegraph explore asks a codebase: one call returns the most relevant beads verbatim — description, notes, comments, close reason — plus each hit's relationship neighborhood, under an output budget.

Fills the gap the stock bd CLI leaves: bd search covers titles, bd query is structured-only, and nothing searches notes, comments, or close reasons — which is where a mature store keeps most of its knowledge. bd memories is indexed too (the plain CLI truncates memory bodies; this returns them whole).

Docs site: https://halaprix.github.io/bd-explore/

$ bd-explore "why did we re-point SYRP status:open"

═══ SYRP-142 [OPEN · P1 · task · updated 2026-08-12]
    Re-point SYRP feed to the v2 oracle
    The v1 oracle staleness window regressed after the chain upgrade…
    COMMENT (ksz 2026-08-11):
    Decision: re-point rather than patch v1 — see close reason on SYRP-118.
    ── neighborhood ──
    blocked by: SYRP-139 — Oracle failover runbook [in_progress]
    child of: SYRP-100 — Oracle migration epic [in_progress]
    mentions: SYRP-118

Key Features

• Deep Verbatim Search: Full Porter-stemmed FTS5 search across title, description, design, acceptance criteria, notes, dated author comments, close reasons, and memories.

• Relational Neighborhood Graphs: Surfaces 1-hop dependencies (blocks, blocked-by, parent-child, supersedes, discovered-from, related), cross-prose mention references, and GitHub issue/PR links (#NNN).

• Transitive Blast Radius: Query transitive dependency chains (--blast <id>) to see blockers, downstream dependents, and epic hierarchy before touching code.

• Built-in Stdio MCP Server: Zero-dependency JSON-RPC 2.0 stdio Model Context Protocol (MCP) server providing the bd_explore tool to modern AI coding assistants.

• Multi-Target Platform Installer: Automated discovery and setup for Claude Code, Gemini CLI, Antigravity IDE, OpenAI Codex, Cursor, and AGENTS.md.

• Beads Persistent Memory Injection: Automatically sets beads memory (bd remember --key bd-explore) so every bd prime session primes agents with bd-explore context.

• Strict Output Budgeting: Output character budget (--budget 24000) prevents context-window blowout in LLM workflows.

• Zero Runtime Dependencies: Pure Python 3.10+ standard library (sqlite3, json, argparse).

Related MCP server: Veridge MCP Server

Installation

Standalone Shell Installer

Install bd-explore into ~/.local/bin and automatically configure detected agent platforms:

# From repository clone
./install.sh

# Standalone uninstall
./install.sh --uninstall

Python Package Installation

# Standard pip install
pip install .

# Editable install for development
pip install -e .

Usage

CLI Search

# Free text search across all fields (porter-stemmed FTS)
bd-explore "why did we re-point SYRP"

# Compose field filters with free text (codegraph-style)
bd-explore "hash refresh status:open type:task priority:1"
bd-explore "swap oracle epic:rpm5"

# Target specific store or force reindex
bd-explore --store ~/Projects/my-project "auth refactor"
bd-explore --rebuild

# Control limits and output budget
bd-explore -n 3 --budget 16000 "database migration"

Supported Filters

Non-filter tokens (e.g. foo:bar) automatically fall through to free-text search. Tip: Quote your search string when it contains spaces, filter colons, or words matching subcommands (e.g. bd-explore "serve refactor").

Transitive Blast Radius

Compute the full transitive dependency graph for any bead:

bd-explore --blast 9o32

Outputs:

• Upstream Blockers: All issues directly or transitively blocking this bead.

• Downstream Blocked: All issues directly or transitively waiting on this bead.

• Epic Ancestry: Direct and ancestor epics.

Stdio MCP Server

bd-explore includes a built-in JSON-RPC 2.0 stdio MCP server for agent integration. It supports both newline-delimited JSON (NDJSON) and HTTP-style Content-Length: header framing.

Run server directly:

bd-explore serve --mcp
# Or with explicit store:
bd-explore serve --mcp --store ~/Projects/my-project

MCP Tool: bd_explore

Exposes the bd_explore tool with schema:

• query (string): Search query string with optional field filters (status:open type:task).

• blast (string): Bead ID to calculate transitive blast radius.

• limit (integer, default 5): Maximum number of seed beads.

• budget (integer, default 24000): Output character budget cap.

• store (string, optional): Explicit store path or repository directory.

Multi-Target Agent Installer

bd-explore install discovers installed AI developer tools, adds MCP configuration, injects marker-fenced agent guidelines, and injects beads persistent memory.

# Interactive setup (prompts for targets and location)
bd-explore install

# Automated non-interactive batch install
bd-explore install --yes

# Install for specific targets and location
bd-explore install --targets claude,gemini,cursor --location global --auto-allow --yes

# Uninstall configurations
bd-explore uninstall --yes

# Print MCP configuration snippet without modifying files
bd-explore print-config claude
bd-explore print-config cursor

Supported Platforms

Marker-Fenced Instructions

Instructions are safely injected with marker fences for clean updates and uninstalls:

<!-- BD_EXPLORE_START -->
## bd-explore

In repositories with a beads store (a `.beads/` directory exists at the repo root), reach for `bd-explore` BEFORE searching raw files or relying only on `bd search`:

- **MCP tool** (when available): `bd_explore` answers questions about beads/issues/decisions/memories verbatim — description, notes, comments, close reason, plus relationship neighborhood under an output budget.
- **Shell** (always works): `bd-explore "<query>"` (e.g. `bd-explore "why did we re-point SYRP status:open"`, `bd-explore --blast <id>`).

If there is no `.beads/` directory, skip bd-explore.
<!-- BD_EXPLORE_END -->

What Gets Indexed

Design Principles

1. Derived and disposable. Reads .beads/issues.jsonl (requires export.auto: true) into a SQLite FTS5 index under ~/.cache/bd-explore/, rebuilt automatically when the export changes. The beads store remains the sole source of truth; delete the cache freely.

2. Staleness is first-class. Every hit is stamped [STATUS · P<n> · type · updated YYYY-MM-DD].

3. Closed beads included by default. History is most of the value; closed hits rank below open ones at equal relevance. Use status:open to narrow.

4. Context window friendly. Strictly enforces output character budgets to fit comfortably into agent conversations.

Architecture

The explore pipeline sits behind one deep module; everything else adapts to it.

              CLI (cli.py)              MCP server (mcp.py)
                   │  thin adapters: args / JSON-RPC  │
                   └──────────────┬───────────────────┘
                                  ▼
                      Explorer (explorer.py)
        explore(query, …) → str   ·   blast(id, …) → str
     owns store discovery, index freshness, connection
       lifetime, defaults/clamping, canonical errors
                   ┌──────────────┴───────────────────┐
                   ▼                                  ▼
          index.py (SQLite FTS5,             search.py (BM25 search,
          mention mining, cache)             hydrate → pure render)

• explorer.py — the only interface callers need: explore() / blast() in, formatted text out, ExploreError on failure.

• index.py — parses .beads/issues.jsonl and bd memories into a derived SQLite FTS5 cache, rebuilt atomically when the export changes.

• search.py — BM25 search and query parsing; hydrate() batch-fetches neighborhoods and titles (two queries total), render() is pure and owns all budget/truncation logic.

• installer/ — multi-target platform adapters behind a common install/uninstall seam.

Domain vocabulary lives in CONTEXT.md; repo conventions in CLAUDE.md.

Development

# Run the full test suite (stdlib unittest — no test dependencies either)
PYTHONPATH=src python3 -m unittest discover tests -v

# Run one module / one case
PYTHONPATH=src python3 -m unittest tests.test_explorer
PYTHONPATH=src python3 -m unittest tests.test_render.TestRenderPure

# Editable install
pip install -e .

CI runs the suite on Linux and macOS across Python 3.10–3.14. See CHANGELOG.md for release history.

Requirements

• Python 3.10+

• SQLite with FTS5 virtual table support (standard in official CPython distributions)

License

MIT License. See LICENSE for details.