Most AI coding assistants only see the file you have open. That’s fine for autocomplete, but useless when you ask “How does the auth service validate tokens?” and the answer lives in a different repo, a Confluence page, or a Jira ticket from three months ago.

This recipe sets up a shared Context Harness instance that gives every engineer’s AI assistant access to your entire engineering org’s knowledge — across repos, issue trackers, and internal docs.

What you’ll build

┌──────────────┐ ┌──────────────┐ ┌──────────────┐ ┌──────────────┐
│ git:platform │ │  git:infra   │ │ s3:runbooks  │ │ script:jira  │
└──────┬───────┘ └──────┬───────┘ └──────┬───────┘ └──────┬───────┘
       │                │                │                │
       └────────────────┴────────────────┴────────────────┘
                                │
                    ┌───────────▼───────────┐
                    │  ctx serve mcp        │
                    │  :7331/mcp            │
                    └───────────┬───────────┘
                                │
          ┌─────────────────────┼─────────────────────┐
          │                     │                     │
    ┌─────▼─────┐        ┌─────▼─────┐        ┌─────▼─────┐
    │  Alice's  │        │  Bob's    │        │ Carol's   │
    │  Cursor   │        │  Cursor   │        │  Claude   │
    └───────────┘        └───────────┘        └───────────┘

One server. Every engineer’s AI gets the same context. No per-person setup beyond a one-line MCP config.

The config

Create a dedicated directory for your team’s context:

mkdir -p ~/team-context/config ~/team-context/connectors

~/team-context/config/ctx.toml:

[db]
path = "./data/ctx.sqlite"

[chunking]
max_tokens = 700
overlap_tokens = 80

[embedding]
provider = "openai"
model = "text-embedding-3-small"
dims = 1536
batch_size = 64

[retrieval]
final_limit = 15
hybrid_alpha = 0.6

[server]
bind = "127.0.0.1:7331"

# ── Source code repos ─────────────────────────────────────────

[connectors.git.platform]
url = "https://github.com/acme/platform.git"
branch = "main"
include_globs = ["docs/**/*.md", "src/**/*.rs", "README.md", "CHANGELOG.md", "ADR/**/*.md"]
exclude_globs = ["**/target/**"]
shallow = true

[connectors.git.auth-service]
url = "https://github.com/acme/auth-service.git"
branch = "main"
include_globs = ["src/**/*.go", "docs/**/*.md", "README.md"]
shallow = true

[connectors.git.frontend]
url = "https://github.com/acme/web-app.git"
branch = "main"
include_globs = ["src/**/*.tsx", "docs/**/*.md", "README.md"]
exclude_globs = ["**/node_modules/**", "**/__tests__/**"]
shallow = true

# ── Runbooks from S3 ─────────────────────────────────────────

[connectors.s3.runbooks]
bucket = "acme-engineering"
prefix = "runbooks/"
region = "us-east-1"
include_globs = ["**/*.md"]

# ── Jira issues ──────────────────────────────────────────────

[connectors.script.jira]
path = "connectors/jira.lua"
timeout = 120
url = "https://acme.atlassian.net"
project = "ENG"
api_token = "${JIRA_API_TOKEN}"
max_results = 500

# ── GitHub Issues ────────────────────────────────────────────

[connectors.script.github-issues]
path = "connectors/github-issues.lua"
owner = "acme"
repo = "platform"
token = "${GITHUB_TOKEN}"
state = "all"
max_pages = 10

# ── Agents ───────────────────────────────────────────────────

[agents.inline.code-reviewer]
description = "Reviews code against project conventions and patterns"
tools = ["search", "get"]
system_prompt = """
You are a senior code reviewer for the Acme engineering team. When
reviewing code:
1. Search for relevant coding conventions, patterns, and ADRs
2. Cite specific documents when suggesting changes
3. Focus on correctness, consistency with existing patterns, and
   potential issues — not style nitpicks
"""

[agents.inline.oncall]
description = "Helps triage production incidents using runbooks and past issues"
tools = ["search", "get"]
system_prompt = """
You are an on-call assistant for the Acme engineering team. When an
incident is reported:
1. Search runbooks for relevant procedures
2. Search past Jira issues for similar incidents
3. Provide step-by-step remediation guidance
4. Cite runbook titles and issue numbers
Be systematic: gather context, identify the issue, recommend actions.
"""

Initial setup

cd ~/team-context

# Initialize the database
ctx init

# Sync all sources (runs in parallel)
ctx sync all

# Generate embeddings for hybrid search
ctx embed pending

# Start the server
ctx serve mcp

The first sync clones repos and fetches all issues — expect 2-5 minutes depending on repo sizes. Subsequent syncs are incremental and much faster.

Connect every engineer

Each engineer adds one file to their Cursor workspace — .cursor/mcp.json:

{
  "mcpServers": {
    "acme-context": {
      "url": "http://127.0.0.1:7331/mcp"
    }
  }
}

Commit this to each repo so everyone gets it automatically. If the server runs on a shared machine, use that machine’s IP instead of 127.0.0.1.

What engineers can do now

Find code patterns across repos:

Get context from issues:

Use runbooks during incidents:

Review code with team conventions:

Keep it fresh

Add a cron job to sync every 2 hours:

# crontab -e
0 */2 * * * cd ~/team-context && ctx sync all && ctx embed pending >> /var/log/ctx-sync.log 2>&1

Or run it as a systemd service that syncs on a timer:

# /etc/systemd/system/ctx-sync.timer
[Unit]
Description=Context Harness sync timer

[Timer]
OnCalendar=*:0/120
Persistent=true

[Install]
WantedBy=timers.target

Scaling tips

The payoff

Before: engineer asks AI about auth, gets a generic answer about JWT. After: engineer asks AI about auth, gets the team’s actual auth implementation with links to the relevant code, ADRs, and related Jira issues.

Context Harness turns your AI assistant from a general-purpose tool into a team member that actually knows your codebase.