Discover MCPs & agents
Loading MCPs and agents…
Loading MCPs and agents…
Rust CLI tools for AI-assisted development. AST-aware code search with tree-sitter + FTS5, and a YAML-configurable output compressor with 28 built-in pipelines. >90% token savings. Part of Gobby, works standalone.
From the repo.
Rust CLI tools for AI-assisted development.
Code search, symbol navigation, hook dispatch, and multimodal research vaults — all from the terminal.
This workspace ships three Gobby CLI tools plus a shared library.
| Crate | Binary | Version | Release tag |
|---|---|---|---|
gobby-code | gcode | 1.4.0 | gcode-v1.4.0 |
gobby-core | n/a | 0.7.0 | gobby-core-v0.7.0 |
gobby-hooks | ghook | 0.7.0 | ghook-v0.7.0 |
gobby-wiki | gwiki | 0.7.0 | gwiki-v0.7.0 |
AST-aware code search powered by tree-sitter. Indexes 21 languages plus safe
repo text files into the Gobby PostgreSQL hub, with pg_search BM25 for symbol
lookup, exact indexed grep over repo content chunks, ranked repo-content search
across source/docs/config/scripts, file tree
navigation, and hybrid ranking. When FalkorDB, Qdrant, and an embedding
source are configured, gcode adds graph-aware search, semantic
search, opt-in graph expansion for exact symbol lookup
(gcode search-symbol --with-graph), dependency analysis (callers, usages,
imports, blast-radius, shortest CALLS path), and Rust-owned graph/vector
projection lifecycle. --token-budget trims high-volume reads (search,
usages, blast-radius) to an approximate token cap, and gcode prune
reconciles stale projects and orphaned graph/vector projection state across all
indexed projects. gcode graph clear --project-id <PROJECT_ID> is available for
daemon stale-project graph cleanup without cwd project resolution.
gcode codewiki generates vault-ready hierarchical code documentation
(repo → modules → files, plus architecture/onboarding/hotspots/changes/ownership
pages) from the index, with optional AI prose tiers (--ai-depth), validated
Mermaid architecture diagrams (workspace topology plus runtime/AI-generation
sequence diagrams), and citation-checked grounding against indexed source
spans. Aggregate pages can use the daemon-side agentic/tool-backed generation
route, and gcode codewiki --purge --out <vault> --force clears generated
CodeWiki output before a clean rebuild. gwiki indexes the output as normal
vault documents.
For non-Gobby-managed projects, gcode init installs the bundled gcode skill
for Claude Code, Codex, Droid, Grok, Qwen, and Antigravity CLI. Gobby-managed
projects skip those project-local skill writes because Gobby owns CLI wiring.
Sandbox-tolerant hook dispatcher invoked by host AI CLIs (Claude Code, Codex, Qwen CLI, Droid, Grok, and Antigravity) on lifecycle and tool-use events. Spools envelopes to ~/.gobby/hooks/inbox/ before POSTing to the local Gobby daemon, so the daemon's drain worker can replay any delivery lost to a sandbox FS-read denial, network blip, or daemon restart. You don't usually invoke it directly — Gobby wires it into your AI CLI for you.
Ingests multimodal sources — documents, PDFs, URLs, MediaWiki, git repos,
Wayback snapshots, audio/image/video, and archived Gobby session transcripts
(gwiki sync-sessions) — into a Markdown knowledge vault with frontmatter
provenance and citations, then indexes and searches them with the same hybrid
BM25 + semantic + graph stack as gcode. gwiki search is the bounded agent
retrieval primitive and gwiki ask is a thin bounded-evidence RAG layer over
it, both honoring --token-budget. Compiles vault material into cited briefs,
emits static agent exports (llms.txt, llms-full.txt, graph.jsonld), and
repairs on-disk markdown with gwiki normalize. Multimodal and AI capabilities
degrade gracefully when transcription, vision, or the configured datastores are
unavailable.
gobby-core underpins them all — a small shared-primitives library for project
root walk-up, bootstrap config, daemon URL composition, setup/provisioning
contracts, and datastore client adapters. It is not a standalone tool.
Download from GitHub Releases. Binaries are available for macOS (ARM/x86), Linux (x86/ARM), and Windows (x86/ARM).
# gcode
cargo install gobby-code
# ghook
cargo install gobby-hooks
# gwiki
cargo install gobby-wiki
gcode graph and semantic features are configured at runtime. There are no
Cargo feature flags for FalkorDB, Qdrant, or embeddings support. Gobby-managed
projects read FalkorDB settings from databases.falkordb.*; daemon-independent
setups can use GOBBY_FALKORDB_HOST, GOBBY_FALKORDB_PORT, and
GOBBY_FALKORDB_PASSWORD.
gcode 0.8.0+ uses the migrated Gobby PostgreSQL hub. It asks the local daemon
broker for the hub DSN first. If the daemon is unavailable, it checks fallback
sources in order: GCODE_DATABASE_URL, GOBBY_POSTGRES_DSN,
~/.gobby/gcore.yaml databases.postgres.dsn, then bootstrap database_url.
Bootstrap fallback is valid only when hub_backend: postgres and bootstrap
contains an inline database_url.
For daemon-independent service provisioning, use gcode setup --standalone.
The default setup path is non-destructive. If incompatible code-index state is
already present, rerun with gcode setup --standalone --overwrite-code-index
only when you intend to reset all gcode-owned code-index PostgreSQL,
FalkorDB, and Qdrant projection state.
Graph/vector lifecycle is code-index scoped. FalkorDB clears target-only
code-index labels, and Qdrant clears target-only code_symbols_{project_id};
Gobby memory graph and memory vector collections stay outside this boundary.
Installing from source or crates.io requires Rust 1.88+.
git clone https://github.com/GobbyAI/gobby-cli.git
cd gobby-cli
cargo install --path crates/gcode
cargo install --path crates/ghook
cargo install --path crates/gwiki
cargo build --workspace --no-default-features # Build all tools
cargo nextest run --workspace --no-default-features # Test all tools except doctests
cargo test --doc --workspace --no-default-features # Test doctests
cargo clippy --workspace --no-default-features -- -D warnings # Lint all tools
cargo fmt --all --check # Check formatting
Apache 2.0 — see LICENSE.
Replace {MCP_ENDPOINT_URL} with this MCP’s endpoint URL (from its repo or docs above). No API key — you connect directly.
Tool
OS
Config file: ~/.cursor/mcp.json
{
"mcpServers": {
"mcp-server": {
"url": "{MCP_ENDPOINT_URL}"
}
}
}Paste into mcpServers in the config file. Restart Cursor after saving.
If this MCP is also published on mcpchannel.ai, you can subscribe from Browse and use the gateway config there instead.