DoableClaw Research MCP

An MCP server exposing DoableClaw's AI research agent — one tool per research domain, callable individually against a company, an industry, or a pitch deck. Backed by doableclaw.com.

• Server: https://ikshan-be-prod-6gbup6suoq-el.a.run.app/mcp/research-v4

• Transport: Streamable HTTP

• Auth: Authorization: Bearer <token> — see Getting a token

• Registry entry: io.github.ikshan-ai/doableclaw-mcp

Tools

Each tool runs one research agent standalone. Most take company_name and/or root_url (at least one required); identity is run automatically by any tool that needs it, so you don't call it yourself unless identity is all you want. Every call costs credits from your DoableClaw account — see cost below.

Costs are set from measured median LLM/tool spend per agent across real research runs, including the identity (and, where relevant, product) resolution every standalone call re-runs from scratch — competitive costs what it does because profiling each competitor genuinely re-runs identity/product/people/etc. research against that competitor.

market is the only tool that works with no company at all — pass just industry (optionally geography) for an industry-wide view.

deck takes deck_text (the deck's extracted text) instead of a company.

Every other tool takes:

• company_name — the company's name

• root_url — the company's website

At least one of company_name/root_url is required.

Tool output is JSON (not prose) — the point of calling one of these is the structured research data, not a chat reply.

Related MCP server: Small Business Intelligence MCP

Getting a token

1. Sign in at doableclaw.com, then go to Account → API Access (doableclaw.com/account/api-keys).

2. Enter a label (e.g. "my laptop") and click Create Key. The plaintext token is shown once — copy it immediately, it can't be retrieved again. It looks like dcresearch_xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx.

3. The same page shows your mcp_research_api credit balance and a Buy credits button if you're out.

Revoke a key any time from the same page — an MCP client using a revoked key starts getting 401s immediately.

Prefer the API directly (e.g. for scripting key rotation)? The page is a thin client over:

# List your keys
curl -s https://ikshan-be-prod-6gbup6suoq-el.a.run.app/api/v1/account/mcp/research-v4/tokens \
  -H "Authorization: Bearer <your doableclaw.com session token>"

# Create one
curl -s -X POST https://ikshan-be-prod-6gbup6suoq-el.a.run.app/api/v1/account/mcp/research-v4/tokens \
  -H "Authorization: Bearer <your doableclaw.com session token>" \
  -H "Content-Type: application/json" \
  -d '{"label": "my laptop"}'

# Revoke one
curl -s -X DELETE https://ikshan-be-prod-6gbup6suoq-el.a.run.app/api/v1/account/mcp/research-v4/tokens/<id> \
  -H "Authorization: Bearer <your doableclaw.com session token>"

The "session token" here is your doableclaw.com login JWT, not the dcresearch_... MCP key itself — these endpoints are how the account page manages MCP keys, not something an MCP client calls.

Client setup

Most MCP clients speak stdio, not remote HTTP directly — bridge with mcp-remote:

{
  "mcpServers": {
    "doableclaw-research": {
      "command": "npx",
      "args": [
        "-y", "mcp-remote",
        "https://ikshan-be-prod-6gbup6suoq-el.a.run.app/mcp/research-v4",
        "--header", "Authorization:Bearer ${DOABLECLAW_MCP_TOKEN}"
      ]
    }
  }
}

Set DOABLECLAW_MCP_TOKEN in your environment, or hardcode the token in place of ${DOABLECLAW_MCP_TOKEN} (less safe — avoid committing it anywhere). Restart your client fully after editing its config.

Credits and pricing

Every call is metered against your DoableClaw account's credit balance for the mcp_research_api capability — the same credits system used for other paid features. A call that would exceed your balance is refused before any research runs (nothing is charged). Check your balance and buy more from Account → API Access on doableclaw.com.

Status

v1.0.0 — first public release. synthesis (the full-pipeline join step) and the underlying skill-level primitives are intentionally not exposed as separate tools; the 12 agents above are the supported granularity.

Support

Issues and questions: open a GitHub issue on this repo.