Context Harness
A local-first knowledge and agent runtime for AI tools.
Power your agent
पब्लिशर का इंस्टॉल। PlugMCP एंडपॉइंट नहीं।
Host a copy with our keys and logs.
रजिस्ट्री में अभी कोई पैकेज्ड रिलीज़ नहीं है। रिपॉज़िटरी का README देखें.
रिमोट URL या आपकी मशीन पर कमांड? दोनों इंस्टॉल में क्या अंतर है
PlugMCP पर चलाएँ
A new server you create. It is not this listing's official endpoint. Use the publisher install when you want their URL and their credentials.
लॉन्चContext Harness turns project knowledge into infrastructure AI can actually use. It ingests files, Git repositories, S3, and custom sources into an inspectable local SQLite index, then gives you two execution paths: reusable profiles for clients such as Cursor and Claude, or governed local agents with scoped tools, durable tasks, typed outcomes, checkpoints, and conservative recovery.
.txt) plus PDF, Word (.docx), PowerPoint (.pptx), and Excel (.xlsx) with automatic text extraction when you include those extensions in include_globsctx commandFor a 5-minute walkthrough with copy-paste config, see the Quick Start guide on the docs site.
Pre-built binaries (recommended):
Download the latest release from GitHub Releases:
# macOS (Apple Silicon)
curl -L https://github.com/parallax-labs/context-harness/releases/latest/download/ctx-macos-aarch64.tar.gz | tar xz
sudo mv ctx /usr/local/bin/
# macOS (Intel)
curl -L https://github.com/parallax-labs/context-harness/releases/latest/download/ctx-macos-x86_64.tar.gz | tar xz
sudo mv ctx /usr/local/bin/
# Linux (x86_64)
curl -L https://github.com/parallax-labs/context-harness/releases/latest/download/ctx-linux-x86_64.tar.gz | tar xz
sudo mv ctx /usr/local/bin/
# Linux (aarch64)
curl -L https://github.com/parallax-labs/context-harness/releases/latest/download/ctx-linux-aarch64.tar.gz | tar xz
sudo mv ctx /usr/local/bin/
Windows: download ctx-windows-x86_64.zip from the releases page and add ctx.exe to your PATH.
Nix (NixOS / nix-darwin):
Install straight from the repo flake — no release tarball needed.
From a clone:
# Build (full binary with local embeddings)
nix build .#default
./result/bin/ctx --version
# Or install into your user profile (on $PATH)
nix profile install .#default
Without cloning (flake URL):
nix profile install github:parallax-labs/context-harness#default
The flake provides two packages:
Use nix develop for a development shell. To use Context Harness inside your own flake (NixOS, Home Manager), see the Nix flake guide.
From source:
Local embeddings have no system dependencies; models are downloaded on first use.
cargo build.library not found for -lc++, install the Xcode Command Line Tools: xcode-select --install. If you use Nix, run nix develop first so the shell provides Zig as the C/C++ compiler; then cargo build works.cargo install --path crates/context-harness
cp config/ctx.example.toml config/ctx.toml
# Edit config/ctx.toml with your settings
Config path defaults to ./config/ctx.toml; use --config to override. See configuration reference for all options.
ctx init
ctx sync all # sync all connectors in parallel
ctx sync git:platform # or sync a specific connector
ctx search "your query" # keyword (default)
ctx search "your query" --mode hybrid # keyword + semantic (needs embeddings)
ctx embed pending # backfill embeddings if using local/ollama/openai
ctx serve mcp
Add to .cursor/mcp.json:
{
"mcpServers": {
"context-harness": {
"url": "http://127.0.0.1:7331/mcp"
}
}
}
Profiles give Cursor, Claude, and custom MCP clients a repeatable role without moving the conversation into Context Harness:
ctx profile init code-reviewer
ctx profile validate
ctx profile test code-reviewer
See Profiles for inline TOML, Lua, Rust, MCP, and REST examples.
Start from the generic examples/local-agents setup or
use the reusable context-harness-agents skill
to configure a project. Then validate and run it from the workspace:
ctx tool bindings validate
ctx agent validate
ctx agent run project-researcher "Summarize the architecture and cite the relevant files"
See the local agents guide for explicit model, connector, agent, tool, permission, history, inspection, and resume configuration. The agent setup skill guide covers installation, invocation, customization, generated files, and validation.
Sources → Ingest + Index → Canonical SQLite Knowledge → CLI / MCP Profiles / Governed Agents
DocumentFull reference: CLI docs.
The server exposes an MCP Streamable HTTP endpoint and REST endpoints. See MCP server reference for details. REST responses follow the schemas in docs/SCHEMAS.md.
Errors follow a consistent format:
{
"error": {
"code": "not_found",
"message": "document not found: abc-123"
}
}
All connector types support named instances — configure multiple of each. Full reference: Built-in connectors.
Scans a local directory. Supported formats: plain text (e.g. .md, .txt, .rs) are always ingested as UTF-8. To also index PDF, Word (.docx), PowerPoint (.pptx), and Excel (.xlsx), add those extensions to include_globs (e.g. "**/*.pdf", "**/*.docx"); they are read as binary and extracted automatically.
[connectors.filesystem.docs]
root = "./docs"
include_globs = ["**/*.md", "**/*.txt", "**/*.pdf", "**/*.docx"]
exclude_globs = ["**/drafts/**"]
follow_symlinks = false
max_extract_bytes = 50_000_000 # skip files larger than this (default: 50MB)
[connectors.filesystem.notes]
root = "./notes"
Ingest documentation from any Git repository — point it at a repo URL and subdirectory:
[connectors.git.platform]
url = "https://github.com/acme/platform.git" # or git@... or local path
branch = "main"
root = "docs/" # scan this subdirectory
include_globs = ["**/*.md", "**/*.rst"]
shallow = true # --depth 1 clone
[connectors.git.auth-service]
url = "https://github.com/acme/auth-service.git"
branch = "main"
Features:
git logIngest documentation from Amazon S3 buckets:
[connectors.s3.runbooks]
bucket = "acme-docs"
prefix = "engineering/runbooks/"
region = "us-east-1"
include_globs = ["**/*.md", "**/*.json"]
# endpoint_url = "http://localhost:9000" # for MinIO / LocalStack
Set AWS_ACCESS_KEY_ID and AWS_SECRET_ACCESS_KEY environment variables.
Features:
LastModified timestamps for incremental syncWrite custom connectors in Lua — no recompilation needed. Scripts have access to HTTP, JSON, environment variables, filesystem, base64, crypto, and logging APIs:
[connectors.script.jira]
path = "connectors/jira.lua"
timeout = 600
url = "https://mycompany.atlassian.net"
api_token = "${JIRA_API_TOKEN}"
project_key = "ENG"
# Scaffold a new connector
ctx connector init jira
# Test it
ctx connector test connectors/jira.lua
# Sync it
ctx sync script:jira
See examples/connectors/github-issues.lua for a complete example.
Install community connectors, tools, and profiles from Git-backed repositories. See Registry overview and Usage guide on the docs site.
ctx registry init --config ./config/ctx.toml
Or during first run, ctx init will offer to install it automatically.
# List all available extensions
ctx registry list --config ./config/ctx.toml
# Search for a connector
ctx registry search jira --config ./config/ctx.toml
# See details
ctx registry info connectors/jira --config ./config/ctx.toml
# Add it to your config (scaffolds the TOML entry with placeholders)
ctx registry add connectors/jira --config ./config/ctx.toml
Tools and profiles from registries are auto-discovered at server startup — they appear in GET /tools/list and GET /profiles/list without explicit config. Connectors need credentials, so they require explicit activation via ctx registry add.
[registries.community]
url = "https://github.com/parallax-labs/ctx-registry.git"
path = "~/.ctx/registries/community"
readonly = true
auto_update = true
[registries.company]
url = "[email protected]:myorg/ctx-extensions.git"
path = "~/.ctx/registries/company"
readonly = true
Registries are resolved with precedence: explicit config > .ctx/ project-local > personal > company > community.
Place a .ctx/ directory in your project root with Lua scripts organized as connectors/<name>/connector.lua, tools/<name>/tool.lua, or profiles/<name>/profile.lua. Legacy agents/<name>/agent.lua profile extensions remain discoverable. They are auto-discovered from any subdirectory.
# Copy to a writable registry for editing
ctx registry override connectors/jira --config ./config/ctx.toml
See the registry docs for the full specification.
Context Harness supports three embedding providers:
[embedding]
provider = "local"
# model = "all-minilm-l6-v2" # default, 384 dims — no config needed
Supported models: all-minilm-l6-v2 (384d), bge-small-en-v1.5 (384d), bge-base-en-v1.5 (768d), bge-large-en-v1.5 (1024d), nomic-embed-text-v1 (768d), nomic-embed-text-v1.5 (768d), multilingual-e5-small (384d), multilingual-e5-base (768d), multilingual-e5-large (1024d).
[embedding]
provider = "ollama"
model = "nomic-embed-text"
dims = 768
# url = "http://localhost:11434" # default
[embedding]
provider = "openai"
model = "text-embedding-3-small"
dims = 1536
Set the OPENAI_API_KEY environment variable before using embedding commands.
Pre-built release binaries are built for six targets. The local embedding provider is included on all targets: primary platforms use fastembed (bundled ORT); Linux musl and macOS Intel use a pure-Rust (tract) backend.
cargo install --path crates/context-harness --no-default-featurescargo build -p context-harness --no-default-features --features local-embeddings-tractSee the configuration docs for full platform notes.
Hybrid search merges keyword (FTS5/BM25) and semantic (cosine similarity) signals with a configurable alpha:
[retrieval]
hybrid_alpha = 0.6 # 0.0 = keyword only, 1.0 = semantic only
Use ctx search "query" --mode hybrid --explain to see score breakdowns. See search reference and docs/HYBRID_SCORING.md for the full specification.
[server]
bind = "127.0.0.1:7331"
For production (Docker, systemd, CI), see Deployment.
See config/ctx.example.toml for a complete example, or the configuration reference on the docs site.
The site also documents the search widget (ctx-search.js) for adding ⌘K search to static sites — see the docs for an example.
If Context Harness is useful to you, consider starring the repo.
AGPL-3.0 — see LICENSE
For commercial licensing inquiries, contact Parallax Labs.
Contributions welcome! Please see CONTRIBUTING.md for guidelines.
Local-first context ingestion and retrieval for AI tools. SQLite + embeddings + MCP server for Cursor & Claude.
टूल
अभी टूल सूची नहीं है। रिमोट सर्वर को पहले हेल्थ चेक के बाद सूची मिलती है।
FAQ
इस पेज पर Cursor, Claude Desktop, Claude Code या VS Code के लिए stdio कॉन्फ़िग कॉपी करें। यह पब्लिशर का इंस्टॉल इस्तेमाल करता है, PlugMCP प्रॉक्सी नहीं।
इस रिकॉर्ड में कोई ज़रूरी एनवायरनमेंट वेरिएबल दर्ज नहीं है।
Context Harness stdio पर लोकल चलता है।
पिछली जाँच 2026-10-28 को हुई। Local-first context ingestion and retrieval for AI tools. SQLite + embeddings + MCP server for Cursor & Claude.