AI Coding MCP (@simplr-ai/dev-mcp)

Connect your own coding agent — Codex, Claude Code, Cursor, or any MCP client — to Simplr's feedback and incident queues, integration guides, guardrails, and live verification tools.

Who changes your code?

Your coding agent runs in your workspace and is the only AI agent that creates or modifies your code and tests. It uses your local repository credentials.

Simplr detects and groups work, supplies context, enforces guardrails, reviews proposed changes, records evidence, and sends approval requests through Telegram or the Notification Centre. Optional OpenAI, Anthropic, or Gemini connections inside Simplr analyze redacted evidence for triage and review; they never change code or initiate a merge or deployment. After Simplr returns an allowed or approved decision, the customer coding agent executes the exact operation with its own credentials.

The dev MCP ships bundled knowledge (so it works offline, with no credentials) plus a few live tools that call the real API with your key to confirm an integration works end-to-end.

Two different MCP servers

• @simplr-ai/dev-mcp (this page) helps you build with Simplr — your coding agent uses it while writing integration code.
• @simplr-ai/mcp is the runtime AI-delegation gateway — it exposes your own registered API endpoints to AI agents at runtime via scoped delegation tokens. Connect it with browser OAuth using Runtime MCP OAuth, so no API key or portal token needs to be pasted into chat or MCP configuration.

Install

Claude Code

claude mcp add simplr-dev -- npx -y @simplr-ai/dev-mcp@2.0.0

Codex

Add the server to the project's Codex configuration and expose the secret only through the agent host environment:

[mcp_servers.simplr]
command = "npx"
args = ["-y", "@simplr-ai/dev-mcp@2.0.0"]
env_vars = ["SIMPLR_API_KEY"]
startup_timeout_sec = 20
tool_timeout_sec = 120
default_tools_approval_mode = "writes"

[mcp_servers.simplr.env]
SIMPLR_API_URL = "https://api.simplr-ai.com"

Set SIMPLR_API_KEY through the trusted agent host's environment or secret manager, then add the MCP server. Do not put the secret in the command or expand it into project configuration:

claude mcp add simplr-dev \
  -e SIMPLR_API_URL=https://api.simplr-ai.com \
  -- npx -y @simplr-ai/dev-mcp@2.0.0

Point SIMPLR_API_URL at the environment you're integrating against — https://api.simplr-ai.com for production, or http://localhost:7002 for local dev.

Cursor / other MCP clients

{
  "mcpServers": {
    "simplr-dev": {
      "command": "npx",
      "args": ["-y", "@simplr-ai/dev-mcp@2.0.0"],
      "env": {
        "SIMPLR_API_URL": "https://api.simplr-ai.com"
      }
    }
  }
}

Inject SIMPLR_API_KEY and optional SIMPLR_USER_TOKEN into the MCP process through the client host's environment or secret manager. Do not commit either value to this JSON file.

Configuration

Env var	Required for	Description
SIMPLR_API_URL	all live tools	API base URL — set it yourself (e.g. https://api.simplr-ai.com or http://localhost:7002). No baked-in default.
SIMPLR_API_KEY	live read/check and feedback tools	Use a secret (sk_*) key for the coding feedback workflow.
SIMPLR_USER_TOKEN	live write tools	Portal token for creating flags/webhooks.

The knowledge tools need no credentials. The live tools need SIMPLR_API_URL (plus a key); the code snippets they return also reference SIMPLR_API_URL, so a single env var points everything at the environment you choose.

Tools

Knowledge (offline, no credentials):

Tool	What it does
list_features	Discover every integrable feature.
get_integration_guide	Full step-by-step guide for one feature (optionally per language).
get_code_snippet	A specific snippet in js, flutter, or curl.
get_api_reference	Raw HTTP surface: method, path, body, response.
get_quickstart	One combined install + initialize block for several features.
search_docs	Keyword search across all bundled feature content.

Live (need credentials):

Tool	What it does
validate_api_key	Confirm your key works; report type + environment.
run_check	Run a real fraud/identity check (email/phone/device).
list_flags	See flag config exactly as the SDK receives it.
create_feature_flag	Scaffold a flag (needs a portal token).
create_webhook	Register an event subscription (needs a portal token).
list_feedback	List actionable product feedback for the key's organization.
get_feedback	Inspect a ticket's context, attachments, comments and history.
claim_feedback	Reserve a ticket and move it to in progress.
update_feedback	Move a claimed ticket, set priority, or attach its pull request.
comment_feedback	Record implementation, test, blocker or handoff progress.
release_feedback	Mark a deployed fix released and notify the identified submitter once.
get_automation_setup_plan	Return the required operator questionnaire, credentials, safety rules and setup order.
inspect_automation_setup	Inspect existing services, rules, runbooks and runs before changing configuration.
apply_automation_setup	Create the confirmed service catalog and starter rules with owner/admin authorization.
run_automation_template	Start a built-in automation and pause sensitive work for approval.
send_automation_test_event	Verify Cloud Monitoring, OpenTelemetry or Sentry ingestion end to end.
connect_incident_ai_provider	Connect OpenAI, Anthropic or Gemini from an MCP-host environment variable with an explicit model and priority.
connect_mongodb_investigation	Connect a read-only MongoDB URI from an MCP-host environment variable and allowlist collections available during triage.
connect_google_cloud_logging	Connect a Logs Viewer service account, project and bounded default filter.
test_incident_connection	Verify AI, MongoDB or Cloud Logging access without returning credentials.
investigate_self_healing_incident	Collect redacted evidence, run provider triage and persist the incident brief.
get_incident_escalation_context	Retrieve everything a human needs for escalation: evidence, attempts, fixes, code/release locations and remaining questions.
request_github_operation	Evaluate an exact review, merge, or deployment against Simplr policy and approvals without executing it.
authorize_external_github_operation	Revalidate the exact operation and return evidence that binds the next customer-agent execution.
record_external_github_operation	Record the result and run ID; Simplr verifies the nonce-bound merge commit or exact new workflow run.
create_test_verification	Start evidence collection after a recorded customer-executed non-production deployment.

Agent-led automation setup

The customer coding agent must begin with get_automation_setup_plan. It first explains the responsibility split, then asks its operator about the organization, scheduler, repositories, environments, owners, telemetry sources, inbound support address, autonomy level and starter templates. It must summarize the exact scope before calling apply_automation_setup with operator_confirmed: true.

Schedule feedback and incident checks

Simplr does not start the customer's coding agent. Create a recurring task in Codex, Claude, your CI system, or another trusted scheduler on a machine that has the repository checked out. A 15-minute interval is a practical starting point.

Use this scheduled task prompt:

Every 15 minutes, use the Simplr MCP tools to run list_feedback and
list_self_healing_incidents. Read the full item, environment, repository and
guardrails. Claim only one item before changing code. Use this coding agent's
own repository credentials to investigate, edit and test.

For an incident, run evaluate_self_healing_action before each proposed code,
database, merge or deployment action. For feedback, load the organization's
coding standards, keep the patch bounded to the claimed item, and submit the
diff to Simplr review. Use request_github_operation for GitHub review, merge and
deployment preflight. Stop when blocked. When approval is required, wait while
Simplr notifies an approver through Telegram or the Notification Centre.

Deploy only to mapped, pre-approved non-production environments, attach test
evidence, and close work only after Simplr telemetry confirms recovery. Never
expose secrets, delete data, or deploy to production autonomously. For an
approved non-production operation, call authorize_external_github_operation
immediately before execution and wait until its returned execute_after time.
Include its nonce in a merge or squash commit message. For deployment, use the configured GitHub actor and input-free workflow
and capture the exact new workflow run ID. Then call
record_external_github_operation with that run ID so Simplr can attest the result
before verification. External rebase merges are blocked because their method
cannot be independently proven after execution.

To verify the connection without changing anything, ask the agent to run validate_api_key, list_feedback, and list_self_healing_incidents. In the Simplr setup wizard, confirm that the secret key has been used, GitHub and a catalog service are linked, self-healing guardrails are enabled, and Telegram or verified email is ready for approvals.

Credential references are restricted to dedicated MCP-host variables named SIMPLR_INCIDENT_AI_KEY_*, SIMPLR_MONGODB_URI_*, or SIMPLR_GCP_SERVICE_ACCOUNT_*. Connection tools cannot read arbitrary host environment variables.

The setup tool requires an owner/admin SIMPLR_USER_TOKEN. Telemetry verification additionally requires an organization secret SIMPLR_API_KEY. Production authority is never inferred: deployments, merges, database changes, authentication, billing and infrastructure actions remain approval-gated; secret changes and data deletion remain blocked.

The operator must separately approve every intelligence connection. AI credentials are encrypted and never returned. MongoDB connections require a read-role database user and explicit collection allowlists. Google Cloud Logging connections require a service account limited to Logs Viewer; Simplr calls the read-only Logging API and never shells out to gcloud. High and critical incidents use all healthy configured AI providers and reconcile their structured results; lower-severity incidents use priority fallback to control latency and cost.

Claiming a self-healing incident queues automatic investigation when an AI provider is available. A diagnosis run can also call the same pipeline explicitly. When an agent records a DEV deployment, Simplr creates release notes containing what happened, what changed, verification, pull request/release/commit locations and remaining questions. It emails verified organization owners/admins outside simulated environments and publishes the same handoff through the Notification Centre. The escalation-context tool reuses that persisted record for portal, Telegram, webhook or future voice escalation.

Automated feedback-to-code workflow

Give the MCP server a secret organization key, then schedule your coding agent to work the feedback queue. The intended lifecycle is:

list_feedback → get_feedback → claim_feedback → code and test → update_feedback / comment_feedback → deploy → release_feedback

The generic update tool cannot mark a ticket done. release_feedback is the explicit terminal action so the final released notification remains tied to an actual deployment. Identified submitters also receive deduplicated emails when the ticket enters in progress, testable, backlog, wishlist, declined, or duplicate states.

Typical flow

1. Ask your agent to "add Simplr feature flags to this app." It calls list_features, then get_integration_guide({ feature: "feature-flags", language: "js" }).
2. The agent writes the integration code from the returned snippets.
3. With a key configured, it calls validate_api_key and list_flags to confirm the flag exists and the SDK can read it.

Covered features

email-verification · phone-verification · device-fingerprinting · behavioral-biometrics · order-fraud · edge-logs · mdm · feature-flags · rum (user-flows) · webhooks · journeys (CI) · ai-gateway