Sniparaproject brain
Menu
Product
Context layerProject context and memory for coding agents without a new chat surface.Project BrainDefinition, boundaries, and the project state an agent starts from.60s demoWatch a cold ACME agent become a briefed agent before the first edit.Proof replaySee what changes when the agent starts from reviewed project state.InstallConnect the hosted MCP endpoint to the agent you already use.Agent surfacesHow the same Project Brain reaches the coding agents teams already use.
Developers
QuickstartConnect an AI coding agent in minutes.Project Brain modelWhat changed, why, impact, next step, safety.Hosted MCPTool contracts for agent clients.API ReferenceProgrammatic context and project intelligence access.GitHub AppRepository sync and PR Answer Packs.Multi-agentCoordinate swarms, claims, state, events and htasks.Snipara SandboxRepeatable execution when validation needs proof.Companion CLIWorkflow continuity for long agent work.
ProofPhilosophyPricingDocs
Sign inStart Free

Continue.dev Integration

Connect Continue Agent mode directly to Hosted MCP. Current Continue configuration is YAML; Streamable HTTP is the primary remote transport, with SSE reserved for compatibility.

Primary path
For LLM agents: use Hosted MCP
Use config.yaml and a locally stored Continue secret. Do not paste an API key into a committed repository file.

Transport Matrix

Use the project slug consistently across all MCP transports. Pick the transport your client actually supports instead of mixing setup stories.

Project slug
HTTP direct
Recommended

Best when the client accepts plain remote MCP over HTTP and you want the simplest setup.

https://api.snipara.com/mcp/YOUR_PROJECT_SLUG
SSE
Advanced

Use only for clients that explicitly require streaming Server-Sent Events transport.

https://api.snipara.com/mcp/YOUR_PROJECT_SLUG/sse
Local stdio
Local only

Use for stdio compatibility, client package development, and advanced tests powered by uvx snipara-mcp. Prefer Hosted MCP when remote MCP is supported.

SNIPARA_PROJECT_SLUG=YOUR_PROJECT_SLUG

1. Store the secret locally

Add SNIPARA_API_KEYthrough Continue's local secrets UI or secret store. The placeholder below resolves at runtime and is safe to keep in the YAML configuration.

2. Add Hosted MCP to config.yaml

name: Snipara Project
version: 1.0.0
schema: v1

mcpServers:
  - name: snipara
    type: streamable-http
    url: https://api.snipara.com/mcp/YOUR_PROJECT_SLUG
    apiKey: ${{ secrets.SNIPARA_API_KEY }}

Add the block to the active Continue config.yaml. Workspace-specific MCP files may also live under .continue/mcpServers/; keep secret values outside Git.

3. Use Agent mode

  1. Reload Continue after saving the configuration.
  2. Open Chat and select Agent mode so MCP tools are available.
  3. Ask Continue to call snipara_context_query for a project question, then inspect the returned source references.

Start with snipara_context_query for source-backed documentation. For a structural code question, use snipara_code_callers when the specialist Code Graph surface is available to the caller.

Transport fallback

Prefer streamable-http at the project-scoped endpoint above. Use the legacy project SSE endpoint only when a pinned Continue version cannot use Streamable HTTP. Localsnipara-mcp remains a stdio compatibility and package-development path, not the normal hosted setup.

Troubleshooting

  • A 401 means the local secret is missing, invalid, or not authorized for the project.
  • A server that connects but exposes no tools is usually using Chat mode instead of Agent mode, or Continue has not reloaded the YAML.
  • Do not translate this example back to Continue's legacy JSON configuration shape.

Related guides