Codedb
Open-source MCP server that answers AI agent code queries from five in-memory indexes — sub-millisecond code intelligence
If your agent greps its way through a monorepo, codedb is one of the cheapest latency and token wins you can install. The 44x–1,323x query speedups and 107x–1,628x token savings are measured on named repos with a stated Apple M4 Pro, and 2.9s cold start on a 7,364-file repo is a one-time bill. The catch is architectural rather than functional: you run a local Zig daemon and own cold-start time on very large codebases.
Verified 17m ago · liveness 82/100 · cite: rightaichoice.com/tools/codedb
- AI agent developers who want structured code retrieval over MCP instead of shelling out to ripgrep on every query
- Claude Code, Codex, Cursor or Windsurf users working in monorepos and 100MB-plus codebases
- Engineers burning token budget on raw grep output during long agent sessions
- Developers who need to inspect a public GitHub repo's structure without cloning it
- Anyone not running an MCP-capable coding agent — this is an agent-facing endpoint, not a standalone IDE
- Small repos where IDE or LSP search already returns results instantly
- Users who need a graphical interface rather than JSON-RPC tools
We scan live Reddit threads, YouTube comments, X posts, G2 reviews and other communities — and hand you an honest verdict in under a minute.
- Honest verdict, not marketing
- Real pros & cons from real users
- Attributed quotes with receipts
3 free scans · no card needed
Skip codedb if your agent already gets instant structural answers from LSP or IDE search in a small repo, if you need a graphical interface, or if you won't run a local MCP daemon that owns its own index state.
Codedb itself is MIT-licensed and free to run; the paid neighbour is CodeDB Pro, the local toolchain with a persistent daemon, batching and higher-leverage edits, priced at Custom and sold alongside the zig* toolchain via codegraff.com installers. Budget-wise, codedb's real cost is the machine time it saves: it replaces repeated ripgrep calls measured at ~59ms and ~1KB of raw output each with sub-millisecond indexed answers.
In short
Codedb — Open-source MCP server that answers AI agent code queries from five in-memory indexes — sub-millisecond code intelligence. Best for AI agent developers who want structured code retrieval over MCP instead of shelling out to ripgrep on every query, Claude Code, Codex, Cursor or Windsurf users working in monorepos and 100MB-plus codebases, Engineers burning token budget on raw grep output during long agent sessions. Free to use.
What's new in Codedb
Checked 5 days agoAcross the latest 4 updates: 4 feature updates.
Codegraff v0.2.24: exact string edits no longer require op=str_replace
Passing old_string and new_string without an op now runs a unique exact replace instead of failing with a no-content-provided error.
Codegraff v0.2.23: memo caches scoped to local OS user
Memo caches are now private to your local OS user and reject unsafe file links; cleared migrated caches stay cleared.
Codegraff v0.2.22: read and edit errors name the actual failure
Failures now report file not found, permission denied, path too long or is a directory with the resolved absolute path, and a daemon-crashing panic on directory reads was fixed.
Codegraff v0.2.21: repeated edits stop re-mapping files from disk
The edit path snapshots from the stat-validated content cache and retains files up to 8 MiB inside a 32 MiB budget, measuring 3MB edits at 8.5ms then about 4ms.
What people actually say about Codedb — is it worth it?
We ran a structured research pass across product reviews, community discussions, and post-purchase forum threads to surface the patterns vendors won't publish themselves. Below: the recurring strengths, the hidden costs people mention most, and the cohort that consistently regrets adopting this tool.
24 mentions across 4 sources (Hacker News, YouTube, Bluesky, GitHub) · researched Jul 15, 2026.
Average across the 4 sources that answered — each source counts once, not each post.
- +Sub-millisecond queries (0.05ms outline, 4ns word lookup).
- +1,000x speedup over ripgrep/shell tools per the docs.
- +Reduces token consumption by over 1,000x.
- +Cold start indexes 56K lines in under 50ms.
- +Open-source MIT license — free to self-host.
- −Too few real user reviews to gauge reliability.
- −One user reported speed feels suspiciously fast.
- −Only useful if your AI agent supports MCP.
- −No official support forums or community chat.
- −Pro pricing unclear from available data.
- • Pro tier pricing and features are not clearly documented in available data.
Viability Score
How well maintained and how widely used is Codedb? Built from what the vendor actually publishes (docs, changelog, tutorials, integrations, pricing), whether the site is live, and how much real users discuss it. How we calculate this
Last calculated: October 2026
How we score →Key Features
- 16 MCP tools: tree, outline, symbol, search, word, hot, deps, read, edit, changes, status, snapshot
- codedb_remote queries any public GitHub repo over MCP without cloning
- codedb_bundle returns full codebase snapshots
- codedb_projects manages multi-project agent work from one index endpoint
- codedb_index triggers on-demand re-indexing
- Serves queries over MCP JSON-RPC 2.0 on stdio or REST on port 7719
- Five in-memory indexes: outlines, inverted word, trigram, dependency graph, content cache
- Inverted word index returns exact identifier locations in 4ns
- Trigram index full-text search at 110μs, 5.5x faster than brute force
- Reverse dependency graph tracks import relationships in 0.05ms
- Content cache answers reads with no disk I/O at 0.04ms
- Background watcher polls every 2s, re-indexes a changed file in under 2ms
- v0.2.24: old_string plus new_string runs a unique exact replace without op=str_replace
- v0.2.23: memo caches scoped to the local OS user, reject unsafe file links
- v0.2.22: read/edit errors name file-not-found, permission-denied, path-too-long or directory
About Codedb
codedb is an open-source (MIT) MCP server that gives AI coding agents structural code intelligence from in-memory indexes instead of paying for a ripgrep process, filesystem walk and regex scan on every lookup. On startup it walks the repo once, builds five indexes, and answers queries over MCP JSON-RPC 2.0 on stdio or REST on port 7719. It's aimed at engineers running Claude Code, Codex, Cursor or Windsurf sessions in monorepos and 100MB-plus codebases where repeated grep cost compounds. The five indexes map to different query shapes: outlines of functions, structs and imports (0.05ms), an inverted word index for exact identifier hits (4ns), a trigram index for full-text search (110μs, 5.5x faster than brute force), a reverse dependency graph (0.05ms), and a raw content cache that serves reads with no disk I/O (0.04ms). A background watcher polls every 2 seconds, checking mtimes then content hashes, so a single changed file re-indexes in under 2ms without a full rescan. Agents get 16 tools: the core 12 (tree, outline, symbol, search, word, hot, deps, read, edit, changes, status, snapshot) plus codedb_remote for querying any public GitHub repo without cloning, codedb_bundle for full codebase snapshots, codedb_projects for multi-project work and codedb_index for on-demand re-indexing. Recent releases sharpened the write path: v0.2.21 retains files up to 8 MiB inside a 32 MiB budget so repeated edits stop re-mapping from disk, v0.2.22 made read and edit errors name the actual failure, v0.2.23 scoped memo caches to your local OS user, and v0.2.24 lets you pass old_string and new_string without an op for a unique exact replace. It's 12,853 lines of Zig with 199 unit tests and zero runtime dependencies, and the installer auto-registers it as an MCP server. Benchmarks on an Apple M4 Pro put codedb 44x to 1,323x faster per query than CLI tools and 107x to 1,628x lighter in tokens than raw ripgrep output; a 7,364-file, 128MB repo indexed in 2.9s. Position it as a
Behind the Verdict
We'd reach for codedb the moment agent sessions start feeling slow in a big repo. The cost it targets isn't wall-clock time so much as token budget: ripgrep dumps raw matching lines, and the model then pays to parse them. The vendor's own worked example — editing handleBatch in a 1,184-line file — puts the traditional rg-plus-cat route at 118,504 bytes against 12,833 bytes with codedb and its sibling muonry, roughly a 9x cut on one task. Repeat that hundreds of times per session and the savings stack. Where it bites is setup. You're installing a daemon and owning local index state. Cold start scales with repo size: under 50ms on a 16-file project, ~2.5s on vitess's 5,028 files, 2.9s on a 7,364-file, 128MB repo. Snapshots make later startups instant, so the first run is the one you notice. Pass if you're not running an MCP-capable agent. There's no GUI here — it's a JSON-RPC endpoint, and if your editor's built-in search already returns instantly on a small repo, you'll gain little. Teams that won't run local infrastructure should skip it entirely. The closest alternatives are editor-owned. Cursor built client-side trigram indexes for its own regex search; Codex searches through a sandboxed shell and eats process startup on every call. codedb's pitch is that it serves any MCP agent, so a Claude Code user and a Cursor user get the same indexes. If you're single-editor, all-in on that vendor's tooling, the gap narrows. One genuine differentiator: codedb_remote queries any public GitHub repo's file trees, outlines and code search through codedb.codegraff.com, no clone required. Useful for triage and due diligence on dependencies. The recent changelog is unglamorous but reassuring — error messages that name file-not-found or permission-denied, memo caches scoped to
Researching Codedb? Get your full AI stack in 60 seconds.
Free, no signup — tell us your goal and get tools matched to your budget & existing stack.
Real-world workflow fit
Concrete scenarios for the personas Codedb actually fits — and what changes day-one when you adopt it.
Install codedb with the one-line shell installer, let the initial walk build the five indexes (about 2.9s on a 128MB repo), then ask the agent for an outline of a service and the reverse dependencies of a module before refactoring it.
Outcome: Sub-millisecond indexed answers and structured responses instead of hundreds of ripgrep calls, with the dependency graph answering import-impact questions that previously took a grep of every import statement.
Point the agent at codedb over MCP JSON-RPC on stdio (or REST on port 7719), use codedb_word for exact identifiers and codedb_index for on-demand re-indexing after a branch switch, and route edits through the verifiable edit path.
Outcome: One wire protocol for search, read, edit and dependency queries, with successful replacements carrying verified:true and concurrent-edit conflicts surfacing as retryable errors rather than lost writes.
Call codedb_remote against the public GitHub repo to pull file trees, symbol outlines and code search through the codedb.codegraff.com cloud path, without cloning anything locally.
Outcome: Structural answers about a third-party codebase in the same MCP tool surface the agent already uses, with no local clone or index build.
Use Cases
- Let Claude Code resolve a function definition from the index instead of spawning a shell process per lookup.
- Cut token consumption on identifier searches in large repos by returning structured results instead of raw ripgrep lines.
- Serve structural outlines to an agent during multi-file refactoring sessions.
- Track reverse dependencies automatically so an agent can judge import impact before an edit.
- Query any public GitHub repo's structure without cloning it, via codedb_remote.
- Keep edits atomic and verifiable: pre-rename fence, post-write re-read, verified:true on success.
- Run multi-project agent sessions through codedb_projects without restarting an index per repo.
- Re-index on demand with codedb_index after large branch switches.
Limitations
- Codedb is a repository-intelligence layer, not a coding agent: the docs describe CodeDB as the piece that maps the repo while Graff or another supported coding agent runs, scores and evolves the task.
- Freshness depends on a background watcher polling every 2 seconds, so results reflect the last poll rather than instant on-disk changes — unless you pair it with the CodeDB Pro toolchain, where the shared wire protocol makes an edit queryable immediately.
- Cold start scales linearly with repo size: under 50ms for 16 files, ~80ms for 100 files, ~2.5s for a 5,028-file repo and 2.9s for a 7,364-file, 128MB repo; snapshots make subsequent startups instant.
- There is no graphical interface — everything is MCP/JSON-RPC or REST on port 7719.
- Large inverted-index reads and wide batches are bounded (result data capped at 64 MiB with an explicit batch_output_limit error), and iterative recent-file access is capped by an environmental LRU rather than by capping tracked files.
as of 2026-10-13
Verification history
We have re-verified Codedb 11 times since . Each pass re-reads the vendor's own pages and re-checks every listed field against that evidence; passes where nothing had changed are marked as such.
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
Showing the 6 most recent of 11 verification passes.
Free to cite with attribution — this page re-verifies continuously.
12-month cost
Project the real annual outlay, including the implied monthly cost when only an annual tier is published.
Vendor list price only. Add-on usage, seat overages, and contract minimums are surfaced under Hidden costs & gotchas.
Plans compared
For each published Codedb tier: who it actually fits, and what it adds vs. the previous tier. Cross-reference the cost calculator above for projected annual outlay.
Community
$0/mo
CodeDB Pro
Custom
Ideal for
Teams already running codedb locally who want a persistent daemon, batching and the muonry edit path instead of the 2-second poll cycle.
What this tier adds
Adds a stateful daemon and the paid local zig* toolchain: muonry SIMD search, structural reads and atomic edits at about 0.7ms per op, with edits queryable instantly.
Where the pricing makes sense
The company stage and team size where Codedb's pricing actually pencils out — and where peers do it cheaper.
Codedb itself is MIT-licensed and free to run; the paid neighbour is CodeDB Pro, the local toolchain with a persistent daemon, batching and higher-leverage edits, priced at Custom and sold alongside the zig* toolchain via codegraff.com installers. Budget-wise, codedb's real cost is the machine time it saves: it replaces repeated ripgrep calls measured at ~59ms and ~1KB of raw output each with sub-millisecond indexed answers.
Setup time & first value
How long it actually takes to get something useful out of Codedb — broken out by persona, not the marketing-page minute.
One-line install (`curl -fsSL https://codedb.codegraff.com/install.sh | sh`) drops the binary and auto-registers it as an MCP server, so the agent has the 16 tools immediately. First value depends on the initial index walk: under 50ms for a 16-file repo, ~80ms at 100 files, ~2.5s for 5,028 files and 2.9s for a 7,364-file, 128MB repo. Snapshots make later startups instant.
Switching to or from Codedb
How to bring data in from common predecessors and how to get it back out — written for the switcher, not the buyer.
- →From raw ripgrep in agent shells: keep your agent, point it at codedb's tools and let the index answer outlines, symbols, deps and search instead of grepping.
- →From IDE-only search: add codedb as an MCP server so Claude Code, Codex, Cursor or Windsurf get structural answers outside the editor window.
- ↗To CodeDB Pro: install the zig* toolchain from codegraff.com to add a persistent daemon, batching and the muonry edit path sharing codedb's wire protocol.
- ↗To Graff plus CodeDB: use the codegraff.com installer when you want the self-improving coding harness that runs and scores tasks alongside the retrieval layer.
Integrations
Resources & Guides
Tutorials & Learning
YouTube returned 6 videos for “Codedb”, and we withheld 6: 6 could not be judged, because “Codedb” is a single word that other videos use for other things. We are showing none, because we could not prove any of them are about Codedb.
Official links
Tools that pair well with Codedb
Common stack mates teams adopt alongside Codedb, with the specific reason each pairing earns its keep.
Chrome DevTools MCP
Open-source MCP server that gives coding agents live Chrome DevTools for debugging, traces, and browser automation.
Continue
Open-source AI coding agent for VS Code and JetBrains — free to fork since Cursor acquired Continue in January 2026.
Refact.ai
Open-source autonomous AI coding agent that plans, executes, and deploys tasks inside VS Code and JetBrains IDEs.
Featured Head-to-Head Comparisons
Codedb vs Spider Cloud
If you need instant code understanding for AI agent tools like Claude Code or Cursor in large monorepos, Codedb is free and purpose-built. If your priority is feeding live web data into RAG pipelines or LLM workflows, Spider Cloud's pay-as-you-go scraping API with AI extraction and browser commands is the better fit. They solve different problems — pick based on whether your bottleneck is code search or web data retrieval.
Codedb vs Temporal Ai
Codedb and Temporal AI solve entirely different AI infrastructure problems. If your bottleneck is AI agents spending too many tokens or time searching a large codebase, Codedb's sub-millisecond MCP queries are a must-have. If your challenge is keeping AI agents alive through crashes, retries, and long-running multi-step tasks, Temporal's durable execution is the answer. Buy both if you have both problems; pick Codedb for code intelligence and Temporal for workflow resilience.
Codedb vs Voyage Ai
If you need sub-millisecond code retrieval for AI coding agents and want a free, open-source MCP server, Codedb is your tool. For enterprise RAG pipelines demanding domain-specific embeddings (finance, legal) and rerankers with 32K context, Voyage AI is the clear choice despite opaque pricing.
Alternatives to Codedb
View all
Chrome DevTools MCP
Open-source MCP server that gives coding agents live Chrome DevTools for debugging, traces, and browser automation.
Frequently Asked Questions
Best-of guides
Used Codedb? Help shape our editorial sentiment research.