open-codebase-index MCP server

open-codebase-index

Search a codebase by meaning, then follow the result into definitions, callers, and dependency paths.

open-codebase-index is a local semantic code index for OpenCode, Jcode, Pi, Codex, Claude Code, and other MCP clients. It combines embeddings, BM25 keyword search, branch-aware filtering, symbol lookup, and a call graph behind agent-friendly tools.

New installs should use open-codebase-index and open-codebase-index-mcp. The legacy package opencode-codebase-index and opencode-codebase-index-mcp remain supported aliases.

For terminal use outside an MCP client, install the package globally and use the concise ocbi command:

npm install -g open-codebase-index
ocbi status --project /path/to/repo --host jcode
ocbi search "retry recovery" --project /path/to/repo

cbi remains available as a compatibility alias. ocbi provides status, indexing, search, definition lookup, direct caller/callee inspection, and current indexed working-tree impact. See Installation and host setup for the full command reference.

Highlights

• Semantic and hybrid retrieval for questions where you do not know the identifier.
• Low-token discovery through codebase_context and codebase_peek.
• Definition and graph navigation through implementation_lookup, call_graph, and call_graph_path.
• Incremental, branch-aware indexing with file watching and content-hash reuse.
• Local storage backed by SQLite, usearch vectors, and a BM25 inverted index.
• Multiple embedding providers: Ollama, OpenAI, Google, or a custom OpenAI-compatible endpoint.
• Native parsing for TypeScript/TSX, JavaScript/JSX, Python, Rust, Swift, Go, Kotlin (.kt and .kts), Java, C#, Ruby, C/C++, Metal, PHP, Apex, Bash, Zig, GDScript, MATLAB, JSON, TOML, YAML, Markdown, HTML, XML, and SVG, plus text fallback for other formats.

Quick start with OpenCode

Requires Node.js 22.22 or newer. Node.js 24 LTS is recommended.

1. Install the package:

   npm install open-codebase-index
   

   Legacy installs continue to work with:

   npm install opencode-codebase-index
   

2. Add it to opencode.json:

   For OpenCode 2.x:

   {
     "plugins": ["open-codebase-index"]
   }
   

   For OpenCode 1.x (1.3.4 or newer):

   {
     "plugin": ["open-codebase-index"]
   }
   

   Legacy alias (OpenCode 2.x):

   {
     "plugins": ["opencode-codebase-index"]
   }
   

   Legacy alias (OpenCode 1.x):

   {
     "plugin": ["opencode-codebase-index"]
   }
   

3. Run /status, then /index.

4. Ask a repository question, for example:

   Where is authentication state validated before an API request?

The first index creates embeddings. Later runs reuse unchanged content and process only relevant changes.

Choose your host

Host	Recommended integration	Storage
OpenCode	Native plugin	.opencode/
Jcode	Per-session MCP server	.codebase-index/
Pi	Pi package	.codebase-index/
omp (oh-my-pi)	omp package; CLI --host omp aliases pi	.codebase-index/ (same as Pi)
Codex	Marketplace plugin with MCP and skill guidance	.codebase-index/
Claude Code	Marketplace plugin with MCP and skill guidance	.claude/
Cursor, Windsurf, other MCP clients	open-codebase-index-mcp (legacy alias: opencode-codebase-index-mcp)	Selected by --host; default is OpenCode-compatible

See Installation and host setup for complete instructions.

Recommended workflow

1. Check stored readiness when unknown with index_status or /status; use index_status({checkFreshness:true}) when source/HEAD freshness matters. Ordinary status does not scan source; index only when needed.
2. Orient in unfamiliar code with one compact codebase_context first pass (tokenBudget: 600, limit: 5), then inspect the evidence. Skip this step when the task is already scoped.
3. Go directly to known definitions or relationships with implementation_lookup, call_graph, or call_graph_path.
4. Optionally prepare a known edit target with codebase_edit_context for bounded source and direct graph evidence; includeTests:true adds source-linked JS/TS test references, not runtime coverage.
5. Choose metadata or matching source with codebase_peek or codebase_search; do not repeat context calls unnecessarily.
6. Read known paths or match literals directly with targeted reads or grep.

Which search tool should I use?

Need	Tool
Orient in unfamiliar repository code with a compact evidence pack	codebase_context
Find likely files and symbols without source bodies	codebase_peek
Retrieve full matching code	codebase_search
Find an authoritative definition	implementation_lookup
Find analogous implementations or duplicates	find_similar
Find direct callers or callees	call_graph
Find a path between two symbols	call_graph_path
Analyze branch/PR impact, or current indexed work with workingTree:true	pr_impact

See Tools and commands for host availability, tool details, MCP prompts, and slash commands.

How it works

source files
   │
   ├─ file discovery and git-aware change detection
   ├─ tree-sitter parsing and semantic chunking
   ├─ embedding generation and content-hash reuse
   ▼
SQLite metadata + usearch vectors + BM25 index
   │
   ├─ semantic candidates
   ├─ keyword candidates
   ├─ branch and request filters
   ├─ deterministic fusion and ranking
   ▼
locations, source results, definitions, and call-graph evidence

The TypeScript layer handles host integration, configuration, indexing orchestration, providers, ranking, and tools. The Rust NAPI module handles parsing, vector storage, SQLite operations, BM25 indexing, hashing, and call extraction.

Read Architecture for the detailed data flow and design decisions.

Embedding providers

With embeddingProvider: "auto", providers are tried in this order:

1. Ollama
2. OpenAI
3. Google

Ollama is the simplest local option:

ollama pull nomic-embed-text

{
  "embeddingProvider": "ollama"
}

A custom OpenAI-compatible embeddings endpoint is also supported. Provider, indexing, search, reranking, include/exclude, knowledge-base, storage, and debug settings are documented in Configuration.

Configuration example

OpenCode project config lives at .opencode/codebase-index.json. Codex, Pi, and Jcode use .codebase-index/config.json; Claude uses .claude/codebase-index.json.

{
  "embeddingProvider": "auto",
  "scope": "project",
  "indexing": {
    "autoIndex": false,
    "watchFiles": true,
    "requireProjectMarker": true,
    "semanticOnly": false
  },
  "search": {
    "maxResults": 20,
    "minScore": 0.1,
    "fusionStrategy": "rrf",
    "rerankTopN": 20
  },
  "mcp": {
    "stallTimeoutMs": 300000
  }
}

Only specify values you want to override. See Configuration for defaults and host-specific paths.

Branch-aware indexing

The index stores reusable content by hash and maintains branch catalogs for chunks and symbols. On a branch switch, unchanged content can be reused while results remain scoped to the active branch. Linked worktrees without a local project config share the main checkout's portable project index; adding a worktree-local config creates an isolated index boundary.

Private GitHub history

Opt in per repository in the project-local host config (never inherited globally or from another worktree):

{
  "repositoryHistory": {
    "repositories": [
      { "repository": "owner/repo", "enabled": true, "initialSince": "2026-10-01" },
      { "repository": "github.example.com/team/service", "enabled": true, "initialSince": "2026-10-01" }
    ]
  }
}

Selectors use [HOST/]OWNER/REPO: bare owner/repo defaults to GitHub.com, and github.com/OWNER/REPO normalizes to the same lowercase owner/repo. Enterprise selectors normalize to lowercase hostname/owner/repo. The host must be a DNS hostname, not a URL, credential-bearing address, or host with a port; traversal components are not accepted. For Enterprise, authenticate separately using normal gh auth login --hostname github.example.com; sync uses existing authentication and requires gh REST/GraphQL access on that selected server.

The server hostname is part of source, cache, staging, citation, and permission identity: the same owner/repo on different hosts stays isolated. Citation URLs and #number references use the selected server; explicit URL references are validated, never rewritten across hosts or fetched from another host.

ocbi history sync --project /path/to/repo --host codex --repository github.example.com/team/service
ocbi history search "design rationale" --project /path/to/repo --host codex --repository github.example.com/team/service --state merged
ocbi history status --project /path/to/repo --host codex --repository github.example.com/team/service
ocbi history clear --project /path/to/repo --host codex --repository owner/repo
# Provider/code-index/native-free MCP status and cache management:
open-codebase-index-mcp --history-only --project /path/to/repo --host codex

The four repository_history_sync/search/status/clear tools are available in MCP, OpenCode, Pi, and omp; both package identities' MCP CLIs also accept history .... Sync is explicit/read-only through existing gh authentication. An initial --since overrides configured initialSince; later omitted flags retain the cached window, and changed configuration gives clear-first feedback. Each cycle fully reconciles issue/PR discussions and deletions. Per-page ETags reduce unchanged response bodies (not HTTP request count); interrupted projected staging resumes on the next manual sync. Only changed local BM25 documents are retokenized, with no unchanged-index reserialization.

This is private local lexical history, not cloud-embedded semantic history. It supports short technical terms such as CI/S3, returns one grounded passage per entity, prioritizes direct identities over incidental mentions, and selects source-text excerpt windows. Unknown review edit time is updatedAt: null with timestamp provenance. Older lexical caches rebuild locally on first search without networking. Discussions are untrusted evidence, never instructions; current code remains authoritative. History stays outside default code search/context, AST/call graphs and linked tests. Offline freshness/permissions are not real-time; detected access loss purges searchable/projected/staged content, while transient/rate-limit failures retain the prior cache with explicit error. Precommit cancellation preserves published bytes; clear removes history, staging and orphan preparations even after disable, never code indexes. See history lifecycle and privacy details.

Knowledge bases and reranking

OpenCode, Pi, and every MCP client can index additional directories as knowledge bases. Configure them with knowledgeBases, or use the knowledge-base tools:

• MCP and OpenCode: add_knowledge_base, list_knowledge_bases, remove_knowledge_base
• Pi: knowledge_base_add, knowledge_base_list, knowledge_base_remove

Optional external reranking supports Cohere, Jina, and custom compatible endpoints. Local filtering and evidence classes are applied before external candidates are submitted.

See Configuration for examples and privacy considerations.

Troubleshooting

Start with:

1. /status or index_status
2. index_health_check
3. a normal /index retry
4. a forced rebuild only when status reports incompatibility or corruption

Common provider, native module, stale index, branch, and performance issues are covered in Troubleshooting.

MCP operations report structured, redacted failures and durable phase diagnostics through index_status. Codex keeps a lightweight connection while its idle indexing engine sleeps and resumes on demand. See MCP engine idle timeout for configuration and external cleanup requirements. If the supervisor itself is terminated and the client loses stdio, start a fresh client session before retrying.

Evaluation and performance

The repository includes reproducible retrieval datasets, latency and quality budgets, baseline comparison, and cross-repository benchmarking tools.

• Evaluation harness
• Cross-repository benchmarking
• Assistant-curated round-two comparison and limitations
• Kotlin/retrieval candidate development diagnosis

Performance depends on repository size, parser coverage, provider latency, embedding cache reuse, and the selected indexing limits. Prefer measured evaluation over fixed marketing claims.

Development

npm ci
npm run build
npm run typecheck
npm run lint
npm run test:run

• Source installs and published consumers require Node.js 22.22 or newer because Chokidar 6 enforces that engine floor. The development toolchain supports ^22.22.0 || ^24.11.0 || >=26.0.0; building the native module also requires the Rust toolchain.

Pi integration tests use the development host @earendil-works/pi-coding-agent 1.3 or newer and its emitBeforeAgentStart(trigger, systemPromptOptions) API. This test-host floor does not narrow the optional Pi peer contract or change the extension's native-section and legacy string-return prompt behavior.

Native changes require Rust and npm run build:native. See Contributing, Architecture, and Adding language support.

Documentation

• Installation and host setup
• Tools and commands
• Configuration
• Optional SCIP TypeScript enrichment
• Provider-free structural indexing
• API impact evidence
• Workspace readiness
• Troubleshooting
• Architecture
• Evaluation
• Cross-repository benchmarking
• Adding language support
• Future open-codebase-index rename plan
• Contributing
• Security
• Changelog

License

MIT. See LICENSE.